chore: Release v7.9.7

* Add warning log when local flag evaluation called before flags loaded

When feature_enabled() is called with only_evaluate_locally=True before
flag definitions are fetched, the SDK silently returns None. This adds a
warning log so users can diagnose the issue immediately.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Move cold start warning to only fire for only_evaluate_locally=True

The warning was in _locally_evaluate_flag which runs for all flag
evaluations, including those that fall back to server-side evaluation.
Move it to the caller where only_evaluate_locally is known, so it only
fires when the caller explicitly opted out of the server fallback.

* Narrow cold start warning to only fire when flags were never fetched

Use `is None` instead of `not` to avoid firing when flags are loaded
but empty (401, 402, no personal_api_key), which already have their
own specific error logs.

* add changeset

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

.github/workflows/release.yml

@@ -1,58 +1,249 @@
 name: "Release"
 
 on:
-  push:
-    branches:
-      - master
-    paths:
-      - "posthog/version.py"
+  pull_request:
+    types: [closed]
+    branches: [master]
   workflow_dispatch:
 
+permissions:
+  contents: read
+
+# Concurrency control: only one release process can run at a time
+# This prevents race conditions if multiple PRs with 'release' label merge simultaneously
+concurrency:
+  group: release
+  cancel-in-progress: false
+
 jobs:
-  release:
-    name: Publish release
+  check-release-label:
+    name: Check for release label
     runs-on: ubuntu-latest
-    permissions:
-      contents: write
-      id-token: write
+    # Run when PR with 'release' label is merged to master
+    if: |
+      github.event_name == 'workflow_dispatch' ||
+       (github.event_name == 'pull_request' &&
+        github.event.pull_request.merged == true &&
+        contains(github.event.pull_request.labels.*.name, 'release'))
+    outputs:
+      should-release: ${{ steps.check.outputs.should-release }}
     steps:
-      - name: Checkout the repository
-        uses: actions/checkout@85e6279cec87321a52edac9c87bce653a07cf6c2
+      - name: Checkout repository
+        uses: actions/checkout@v4
         with:
+          ref: master
           fetch-depth: 0
 
+      - name: Check release conditions
+        id: check
+        run: |
+          changeset_count=$(find .sampo/changesets -name '*.md' 2>/dev/null | wc -l)
+          if [ "$changeset_count" -gt 0 ]; then
+            echo "should-release=true" >> "$GITHUB_OUTPUT"
+            echo "Found $changeset_count changeset(s), ready to release"
+          else
+            echo "should-release=false" >> "$GITHUB_OUTPUT"
+            echo "No changesets to release"
+          fi
+
+  notify-approval-needed:
+    name: Notify Slack - Approval Needed
+    needs: check-release-label
+    if: needs.check-release-label.outputs.should-release == 'true'
+    uses: posthog/.github/.github/workflows/notify-approval-needed.yml@main
+    with:
+      slack_channel_id: ${{ vars.SLACK_APPROVALS_CLIENT_LIBRARIES_CHANNEL_ID }}
+      slack_user_group_id: ${{ vars.GROUP_CLIENT_LIBRARIES_SLACK_GROUP_ID }}
+    secrets:
+      slack_bot_token: ${{ secrets.SLACK_CLIENT_LIBRARIES_BOT_TOKEN }}
+      posthog_project_api_key: ${{ secrets.POSTHOG_PROJECT_API_KEY }}
+
+  release:
+    name: Release and publish
+    needs: [check-release-label, notify-approval-needed]
+    runs-on: ubuntu-latest
+    # Use `always()` to ensure the job runs even if notify-approval-needed is skipped,
+    # but still depend on it to access `needs.notify-approval-needed.outputs.slack_ts`
+    if: always() && needs.check-release-label.outputs.should-release == 'true'
+    environment: "Release" # This will require an approval from a maintainer, they are notified in Slack above
+    permissions:
+      contents: write
+      actions: write
+      id-token: write
+    steps:
+      - name: Notify Slack - Approved
+        if: needs.notify-approval-needed.outputs.slack_ts != ''
+        uses: posthog/.github/.github/actions/slack-thread-reply@main
+        with:
+          slack_bot_token: ${{ secrets.SLACK_CLIENT_LIBRARIES_BOT_TOKEN }}
+          slack_channel_id: ${{ vars.SLACK_APPROVALS_CLIENT_LIBRARIES_CHANNEL_ID }}
+          thread_ts: ${{ needs.notify-approval-needed.outputs.slack_ts }}
+          message: "✅ Release approved! Version bump in progress..."
+          emoji_reaction: "white_check_mark"
+
+      - name: Get GitHub App token
+        id: releaser
+        uses: actions/create-github-app-token@v2
+        with:
+          app-id: ${{ secrets.GH_APP_POSTHOG_PYTHON_RELEASER_APP_ID }}
+          private-key: ${{ secrets.GH_APP_POSTHOG_PYTHON_RELEASER_PRIVATE_KEY }}
+
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          ref: master
+          fetch-depth: 0
+          token: ${{ steps.releaser.outputs.token }}
+
       - name: Set up Python
-        uses: actions/setup-python@8d9ed9ac5c53483de85588cdf95a591a75ab9f55
+        uses: actions/setup-python@v5
         with:
           python-version: 3.11.11
 
       - name: Install uv
-        uses: astral-sh/setup-uv@0c5e2b8115b80b4c7c5ddf6ffdd634974642d182 # v5.4.1
+        uses: astral-sh/setup-uv@v5
         with:
-            enable-cache: true
-            pyproject-file: 'pyproject.toml'
-            
-      - name: Detect version
-        run: echo "REPO_VERSION=$(python3 posthog/version.py)" >> $GITHUB_ENV
+          enable-cache: true
+          pyproject-file: "pyproject.toml"
 
-      - name: Prepare for building release
+      - name: Install Rust
+        uses: dtolnay/rust-toolchain@0b1efabc08b657293548b77fb76cc02d26091c7e
+        with:
+          toolchain: 1.91.1
+          components: cargo
+
+      - name: Cache Sampo CLI
+        id: cache-sampo
+        uses: actions/cache@v3
+        with:
+          path: ~/.cargo/bin/sampo
+          key: sampo-${{ runner.os }}-${{ runner.arch }}
+
+      - name: Install Sampo CLI
+        if: steps.cache-sampo.outputs.cache-hit != 'true'
+        run: cargo install sampo
+
+      - name: Install dependencies
         run: uv sync --extra dev
 
-      - name: Push releases to PyPI
-        env:
-          TWINE_USERNAME: __token__
-        run: uv run make release && uv run make release_analytics
+      - name: Configure Git
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "github-actions[bot]@users.noreply.github.com"
 
-      - name: Create GitHub release
+      - name: Prepare release with Sampo
+        id: sampo-release
+        env:
+          GITHUB_TOKEN: ${{ steps.releaser.outputs.token }}
+        run: |
+          sampo release
+          new_version=$(python3 -c "import tomllib; print(tomllib.load(open('pyproject.toml', 'rb'))['project']['version'])")
+          echo "new_version=$new_version" >> "$GITHUB_OUTPUT"
+
+      - name: Sync version to posthog/version.py
+        run: |
+          echo 'VERSION = "${{ steps.sampo-release.outputs.new_version }}"' > posthog/version.py
+
+      - name: Commit release changes
+        id: commit-release
+        env:
+          GITHUB_TOKEN: ${{ steps.releaser.outputs.token }}
+        run: |
+          git add -A
+          if git diff --staged --quiet; then
+            echo "No changes to commit"
+            echo "committed=false" >> "$GITHUB_OUTPUT"
+          else
+            git commit -m "chore: Release v${{ steps.sampo-release.outputs.new_version }}"
+            git push origin master
+            echo "committed=true" >> "$GITHUB_OUTPUT"
+          fi
+
+      # Publishing is done manually (not via `sampo publish`) because we need to
+      # publish both `posthog` and `posthoganalytics` packages to PyPI.
+      # Sampo only knows about the `posthog` package, so we handle both here.
+      # Both packages use PyPI OIDC trusted publishing (no API tokens needed).
+      - name: Build posthog
+        if: steps.commit-release.outputs.committed == 'true'
+        run: uv run make build_release
+
+      - name: Publish posthog to PyPI
+        if: steps.commit-release.outputs.committed == 'true'
+        uses: pypa/gh-action-pypi-publish@release/v1
+
+      # The `posthoganalytics` package is a mirror of `posthog` published under
+      # a different name for backwards compatibility. The make target handles
+      # copying, renaming imports, and building the dist automatically.
+      - name: Build posthoganalytics
+        if: steps.commit-release.outputs.committed == 'true'
+        run: uv run make build_release_analytics
+
+      - name: Publish posthoganalytics to PyPI
+        if: steps.commit-release.outputs.committed == 'true'
+        uses: pypa/gh-action-pypi-publish@release/v1
+
+      # We skip `sampo publish` (which normally creates the tag) because we
+      # need to publish both posthog and posthoganalytics manually, so we
+      # create the tag ourselves.
+      - name: Tag release
+        if: steps.commit-release.outputs.committed == 'true'
+        run: git tag "v${{ steps.sampo-release.outputs.new_version }}"
+
+      - name: Push tags
+        if: steps.commit-release.outputs.committed == 'true'
+        run: git push origin --tags
+
+      - name: Create GitHub Release
+        if: steps.commit-release.outputs.committed == 'true'
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: gh release create "v${{ steps.sampo-release.outputs.new_version }}" --generate-notes
+
+      - name: Dispatch generate-references
+        if: steps.commit-release.outputs.committed == 'true'
         env:
           GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          gh release create "v${{ env.REPO_VERSION }}" \
-            --title "${{ env.REPO_VERSION }}" \
-            --generate-notes
-      
-      - name: Dispatch generate-references for posthog-python
-        env:
-          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          gh workflow run generate-references.yml --ref master
+        run: gh workflow run generate-references.yml --ref master
+
+      # Notify in case of a failure
+      - name: Send failure event to PostHog
+        if: ${{ failure() }}
+        uses: PostHog/posthog-github-action@v0.1
+        with:
+          posthog-token: "${{ secrets.POSTHOG_PROJECT_API_KEY }}"
+          event: "posthog-python-github-release-workflow-failure"
+          properties: >-
+            {
+              "commitSha": "${{ github.sha }}",
+              "jobStatus": "${{ job.status }}",
+              "ref": "${{ github.ref }}",
+              "version": "v${{ steps.sampo-release.outputs.new_version }}"
+            }
+
+      - name: Notify Slack - Failed
+        if: ${{ failure() && needs.notify-approval-needed.outputs.slack_ts != '' }}
+        uses: posthog/.github/.github/actions/slack-thread-reply@main
+        with:
+          slack_bot_token: ${{ secrets.SLACK_CLIENT_LIBRARIES_BOT_TOKEN }}
+          slack_channel_id: ${{ vars.SLACK_APPROVALS_CLIENT_LIBRARIES_CHANNEL_ID }}
+          thread_ts: ${{ needs.notify-approval-needed.outputs.slack_ts }}
+          message: "❌ Failed to release `posthog-python@v${{ steps.sampo-release.outputs.new_version }}`! <https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}|View logs>"
+          emoji_reaction: "x"
+
+  notify-released:
+    name: Notify Slack - Released
+    needs: [check-release-label, notify-approval-needed, release]
+    runs-on: ubuntu-latest
+    if: always() && needs.release.result == 'success' && needs.notify-approval-needed.outputs.slack_ts != ''
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Notify Slack - Released
+        uses: posthog/.github/.github/actions/slack-thread-reply@main
+        with:
+          slack_bot_token: ${{ secrets.SLACK_CLIENT_LIBRARIES_BOT_TOKEN }}
+          slack_channel_id: ${{ vars.SLACK_APPROVALS_CLIENT_LIBRARIES_CHANNEL_ID }}
+          thread_ts: ${{ needs.notify-approval-needed.outputs.slack_ts }}
+          message: "🚀 posthog-python released successfully!"
+          emoji_reaction: "rocket"

.github/workflows/sdk-compliance.yml

@@ -0,0 +1,21 @@
+name: SDK Compliance Tests
+
+permissions:
+  contents: read
+  packages: read
+  pull-requests: write
+
+on:
+  pull_request:
+  push:
+    branches:
+      - master
+
+jobs:
+  compliance:
+    name: PostHog SDK compliance tests
+    uses: PostHog/posthog-sdk-test-harness/.github/workflows/test-sdk-action.yml@main
+    with:
+      adapter-dockerfile: "sdk_compliance_adapter/Dockerfile"
+      adapter-context: "."
+      test-harness-version: "latest"

.sampo/config.toml

@@ -0,0 +1,19 @@
+# Sampo configuration
+version = 1
+
+[git]
+default_branch = "master"
+short_tags = "posthog"    # Tag with v1.2.3 rather than posthog-v1.2.3
+
+[github]
+repository = "posthog/posthog-python"
+
+[changelog]
+# Options for release notes generation.
+# show_commit_hash = true (default)
+# show_acknowledgments = true (default)
+
+[packages]
+# Options for package discovery and filtering.
+# ignore_unpublished = false (default)
+# ignore = ["internal-*", "examples/*"]

CHANGELOG.md

@@ -1,33 +1,116 @@
-# 7.6.0 - 2026-01-12
+# posthog
+
+## 7.9.7 — 2026-03-05
+
+### Patch changes
+
+- [b206669](https://github.com/posthog/posthog-python/commit/b206669bf62c923346ad28881dc4694d933ca424) fix(llma): use distinct_id from outer context if not provided, fix $process_person_profile for context-based identity — Thanks @ethanporcaro for your first contribution 🎉!
+- [a99c7d7](https://github.com/posthog/posthog-python/commit/a99c7d73b1e0ef1f35d856c82ace21237ee253a3) Add warning log for local flag evaluation cold start — Thanks @dmarticus!
+
+## 7.9.6 — 2026-03-02
+
+### Patch changes
+
+- [8d83315](https://github.com/posthog/posthog-python/commit/8d83315b67c21eb9e7d6c17bae27ada98ca2643d) add PROPERTY_OPERATORS constant for match_property — Thanks @dmarticus!
+
+## 7.9.5 — 2026-03-02
+
+### Patch changes
+
+- [830244b](https://github.com/posthog/posthog-python/commit/830244bd409b1992ae2e49610f8f87d2cdfc8096) add semver targeting support to local evaluation — Thanks @dmarticus!
+
+## 7.9.4 — 2026-02-25
+
+### Patch changes
+
+- [a68a6a6](https://github.com/posthog/posthog-python/commit/a68a6a6d045072c88eeee7acac441536919b5954) feat(llma): add `$ai_tokens_source` property ("sdk" or "passthrough") to all `$ai_generation` events to detect when token values are externally overridden via `posthog_properties` — Thanks @carlos-marchal-ph!
+
+## 7.9.3 — 2026-02-18
+
+### Patch changes
+
+- [9f9553a](https://github.com/posthog/posthog-python/commit/9f9553a420d22e5e6435b775993f61a059280c2a) Fix posthoganalytics release, previously broken — Thanks @rafaeelaudibert!
+
+## 7.9.2 — 2026-02-18
+
+### Patch changes
+
+- [f1dc4d7](https://github.com/posthog/posthog-python/commit/f1dc4d73914712983a7f715ee4fe1b70e66e770a) Add sampo to the project — Thanks @rafaeelaudibert!
+
+## 7.9.1 - 2026-02-17
+
+fix(llma): make prompt fetches deterministic by requiring project_api_key and sending it as token query param
+
+## 7.9.0 - 2026-02-17
+
+feat: Support device_id as bucketing identifier for local evaluation
+
+## 7.8.6 - 2026-02-09
+
+fix: limit collections scanning in code variables
+
+## 7.8.5 - 2026-02-09
+
+fix: further optimize code variables pattern matching
+
+## 7.8.4 - 2026-02-09
+
+fix: do not pattern match long values in code variables
+
+## 7.8.3 - 2026-02-06
+
+fix: openAI input image sanitization
+
+## 7.8.2 - 2026-02-04
+
+fix(llma): fix prompts default url
+
+## 7.8.1 - 2026-02-03
+
+fix(llma): small fixes for prompt management
+
+## 7.8.0 - 2026-01-28
+
+feat(llma): add prompt management
+
+Adds the Prompt Management feature. At the time of release, this feature is in a closed alpha.
+
+## 7.7.0 - 2026-01-15
+
+feat(ai): Add OpenAI Agents SDK integration
+
+Automatic tracing for agent workflows, handoffs, tool calls, guardrails, and custom spans. Includes `$ai_total_tokens`, `$ai_error_type` categorization, and `$ai_framework` property.
+
+## 7.6.0 - 2026-01-12
 
 feat: add device_id to flags request payload
 
 Add device_id parameter to all feature flag methods, allowing the server to track device identifiers for flag evaluation. The device_id can be passed explicitly or set via context using `set_context_device_id()`.
 
-# 7.5.1 - 2026-01-07
+## 7.5.1 - 2026-01-07
 
 fix: avoid return from finally block to fix Python 3.14 SyntaxWarning (#361) - thanks @jodal
 
-# 7.5.0 - 2026-01-06
+## 7.5.0 - 2026-01-06
 
 feat: Capture Langchain, OpenAI and Anthropic errors as exceptions (if exception autocapture is enabled)
 feat: Add reference to exception in LLMA trace and span events
 
-# 7.4.3 - 2026-01-02
+## 7.4.3 - 2026-01-02
 
 Fixes cache creation cost for Langchain with Anthropic
 
-# 7.4.2 - 2025-12-22
+## 7.4.2 - 2025-12-22
 
 feat: add `in_app_modules` option to control code variables capturing
 
-# 7.4.1 - 2025-12-19
+## 7.4.1 - 2025-12-19
 
 fix: extract model from response for OpenAI stored prompts
 
 When using OpenAI stored prompts, the model is defined in the OpenAI dashboard rather than passed in the API request. This fix adds a fallback to extract the model from the response object when not provided in kwargs, ensuring generations show up with the correct model and enabling cost calculations.
 
-# 7.4.0 - 2025-12-16
+## 7.4.0 - 2025-12-16
 
 feat: Add automatic retries for feature flag requests
 
@@ -39,32 +122,32 @@ Feature flag API requests now automatically retry on transient failures:
 
 Rate limit (429) and quota (402) errors are not retried.
 
-# 7.3.1 - 2025-12-06
+## 7.3.1 - 2025-12-06
 
 fix: remove unused $exception_message and $exception_type
 
-# 7.3.0 - 2025-12-05
+## 7.3.0 - 2025-12-05
 
 feat: improve code variables capture masking
 
-# 7.2.0 - 2025-12-01
+## 7.2.0 - 2025-12-01
 
 feat: add $feature_flag_evaluated_at properties to $feature_flag_called events
 
-# 7.1.0 - 2025-11-26
+## 7.1.0 - 2025-11-26
 
 Add support for the async version of Gemini.
 
-# 7.0.2 - 2025-11-18
+## 7.0.2 - 2025-11-18
 
 Add support for Python 3.14.
 Projects upgrading to Python 3.14 should ensure any Pydantic models passed into the SDK use Pydantic v2, as Pydantic v1 is not compatible with Python 3.14.
 
-# 7.0.1 - 2025-11-15
+## 7.0.1 - 2025-11-15
 
 Try to use repr() when formatting code variables
 
-# 7.0.0 - 2025-11-11
+## 7.0.0 - 2025-11-11
 
 NB Python 3.9 is no longer supported
 
@@ -78,155 +161,155 @@ NB Python 3.9 is no longer supported
   - langchain-community: 0.3.29 → 0.4.1
   - langgraph: 0.6.6 → 1.0.2
 
-# 6.9.3 - 2025-11-10
+## 6.9.3 - 2025-11-10
 
 - feat(ph-ai): PostHog properties dict in GenerationMetadata
 
-# 6.9.2 - 2025-11-10
+## 6.9.2 - 2025-11-10
 
 - fix(llma): fix cache token double subtraction in Langchain for non-Anthropic providers causing negative costs
 
-# 6.9.1 - 2025-11-07
+## 6.9.1 - 2025-11-07
 
 - fix(error-tracking): pass code variables config from init to client
 
-# 6.9.0 - 2025-11-06
+## 6.9.0 - 2025-11-06
 
 - feat(error-tracking): add local variables capture
 
-# 6.8.0 - 2025-11-03
+## 6.8.0 - 2025-11-03
 
 - feat(llma): send web search calls to be used for LLM cost calculations
 
-# 6.7.14 - 2025-11-03
+## 6.7.14 - 2025-11-03
 
 - fix(django): Handle request.user access in async middleware context to prevent SynchronousOnlyOperation errors in Django 5+ (fixes #355)
 - test(django): Add Django 5 integration test suite with real ASGI application testing async middleware behavior
 
-# 6.7.13 - 2025-11-02
+## 6.7.13 - 2025-11-02
 
 - fix(llma): cache cost calculation in the LangChain callback
 
-# 6.7.12 - 2025-11-02
+## 6.7.12 - 2025-11-02
 
 - fix(django): Restore process_exception method to capture view and downstream middleware exceptions (fixes #329)
 - fix(ai/langchain): Add LangChain 1.0+ compatibility for CallbackHandler imports (fixes #362)
 
-# 6.7.11 - 2025-10-28
+## 6.7.11 - 2025-10-28
 
 - feat(ai): Add `$ai_framework` property for framework integrations (e.g. LangChain)
 
-# 6.7.10 - 2025-10-24
+## 6.7.10 - 2025-10-24
 
 - fix(django): Make middleware truly hybrid - compatible with both sync (WSGI) and async (ASGI) Django stacks without breaking sync-only deployments
 
-# 6.7.9 - 2025-10-22
+## 6.7.9 - 2025-10-22
 
 - fix(flags): multi-condition flags with static cohorts returning wrong variants
 
-# 6.7.8 - 2025-10-16
+## 6.7.8 - 2025-10-16
 
 - fix(llma): missing async for OpenAI's streaming implementation
 
-# 6.7.7 - 2025-10-14
+## 6.7.7 - 2025-10-14
 
 - fix: remove deprecated attribute $exception_personURL from exception events
 
-# 6.7.6 - 2025-09-16
+## 6.7.6 - 2025-09-16
 
 - fix: don't sort condition sets with variant overrides to the top
 - fix: Prevent core Client methods from raising exceptions
 
-# 6.7.5 - 2025-09-16
+## 6.7.5 - 2025-09-16
 
 - feat: Django middleware now supports async request handling.
 
-# 6.7.4 - 2025-09-05
+## 6.7.4 - 2025-09-05
 
 - fix: Missing system prompts for some providers
 
-# 6.7.3 - 2025-09-04
+## 6.7.3 - 2025-09-04
 
 - fix: missing usage tokens in Gemini
 
-# 6.7.2 - 2025-09-03
+## 6.7.2 - 2025-09-03
 
 - fix: tool call results in streaming providers
 
-# 6.7.1 - 2025-09-01
+## 6.7.1 - 2025-09-01
 
 - fix: Add base64 inline image sanitization
 
-# 6.7.0 - 2025-08-26
+## 6.7.0 - 2025-08-26
 
 - feat: Add support for feature flag dependencies
 
-# 6.6.1 - 2025-08-21
+## 6.6.1 - 2025-08-21
 
 - fix: Prevent `NoneType` error when `group_properties` is `None`
 
-# 6.6.0 - 2025-08-15
+## 6.6.0 - 2025-08-15
 
 - feat: Add `flag_keys_to_evaluate` parameter to optimize feature flag evaluation performance by only evaluating specified flags
 - feat: Add `flag_keys_filter` option to `send_feature_flags` for selective flag evaluation in capture events
 
-# 6.5.0 - 2025-08-08
+## 6.5.0 - 2025-08-08
 
 - feat: Add `$context_tags` to an event to know which properties were included as tags
 
-# 6.4.1 - 2025-08-06
+## 6.4.1 - 2025-08-06
 
 - fix: Always pass project API key in `remote_config` requests for deterministic project routing
 
-# 6.4.0 - 2025-08-05
+## 6.4.0 - 2025-08-05
 
 - feat: support Vertex AI for Gemini
 
-# 6.3.4 - 2025-08-04
+## 6.3.4 - 2025-08-04
 
 - fix: set `$ai_tools` for all providers and `$ai_output_choices` for all non-streaming provider flows properly
 
-# 6.3.3 - 2025-08-01
+## 6.3.3 - 2025-08-01
 
 - fix: `get_feature_flag_result` now correctly returns FeatureFlagResult when payload is empty string instead of None
 
-# 6.3.2 - 2025-07-31
+## 6.3.2 - 2025-07-31
 
 - fix: Anthropic's tool calls are now handled properly
 
-# 6.3.0 - 2025-07-22
+## 6.3.0 - 2025-07-22
 
 - feat: Enhanced `send_feature_flags` parameter to accept `SendFeatureFlagsOptions` object for declarative control over local/remote evaluation and custom properties
 
-# 6.2.1 - 2025-07-21
+## 6.2.1 - 2025-07-21
 
 - feat: make `posthog_client` an optional argument in PostHog AI providers wrappers (`posthog.ai.*`), intuitively using the default client as the default
 
-# 6.1.1 - 2025-07-16
+## 6.1.1 - 2025-07-16
 
 - fix: correctly capture exceptions processed by Django from views or middleware
 
-# 6.1.0 - 2025-07-10
+## 6.1.0 - 2025-07-10
 
 - feat: decouple feature flag local evaluation from personal API keys; support decrypting remote config payloads without relying on the feature flags poller
 
-# 6.0.4 - 2025-07-09
+## 6.0.4 - 2025-07-09
 
 - fix: add POSTHOG_MW_CLIENT setting to django middleware, to support custom clients for exception capture.
 
-# 6.0.3 - 2025-07-07
+## 6.0.3 - 2025-07-07
 
 - feat: add a feature flag evaluation cache (local storage or redis) to support returning flag evaluations when the service is down
 
-# 6.0.2 - 2025-07-02
+## 6.0.2 - 2025-07-02
 
 - fix: send_feature_flags changed to default to false in `Client::capture_exception`
 
-# 6.0.1
+## 6.0.1
 
 - fix: response `$process_person_profile` property when passed to capture
 
-# 6.0.0
+## 6.0.0
 
 This release contains a number of major breaking changes:
 
@@ -253,15 +336,15 @@ with posthog.new_context():
 
 Generally, arguments are now appropriately typed, and docstrings have been updated. If something is unclear, please open an issue, or submit a PR!
 
-# 5.4.0 - 2025-06-20
+## 5.4.0 - 2025-06-20
 
 - feat: add support to session_id context on page method
 
-# 5.3.0 - 2025-06-19
+## 5.3.0 - 2025-06-19
 
 - fix: safely handle exception values
 
-# 5.2.0 - 2025-06-19
+## 5.2.0 - 2025-06-19
 
 - feat: construct artificial stack traces if no traceback is available on a captured exception
 

Makefile

@@ -5,12 +5,26 @@ test:
 	coverage run -m pytest
 	coverage report
 
-release:
+build_release:
 	rm -rf dist/*
 	python setup.py sdist bdist_wheel
-	twine upload dist/*
 
-release_analytics:
+# Builds the `posthoganalytics` PyPI package, which is a mirror of `posthog`
+# published under a different name for internal use by posthog/posthog.
+#
+# The process works in three phases:
+#   1. posthog -> posthoganalytics: Copy the source, rewrite all imports,
+#      remove the original posthog/ dir, and build the dist.
+#   2. posthoganalytics -> posthog: Reverse the import rewrites, copy
+#      everything back into posthog/, and clean up.
+#   3. Restore pyproject.toml from backup (setup_analytics.py modifies it).
+#
+# This ensures the working tree is left in the same state it started in.
+#
+# NOTE: This target clears dist/ before building. In the release workflow,
+# `build_release` (posthog) must be published BEFORE running this target,
+# otherwise the posthog dist artifacts will be lost.
+build_release_analytics:
 	rm -rf dist
 	rm -rf build
 	rm -rf posthoganalytics
@@ -21,7 +35,6 @@ release_analytics:
 	find ./posthoganalytics -name "*.bak" -delete
 	rm -rf posthog
 	python setup_analytics.py sdist bdist_wheel
-	twine upload dist/*
 	mkdir posthog
 	find ./posthoganalytics -type f -name "*.py" -exec sed -i.bak -e 's/from posthoganalytics /from posthog /g' {} \;
 	find ./posthoganalytics -type f -name "*.py" -exec sed -i.bak -e 's/from posthoganalytics\./from posthog\./g' {} \;
@@ -54,4 +67,4 @@ prep_local:
 	@echo "Local copy created at ../posthog-python-local"
 	@echo "Install with: pip install -e ../posthog-python-local"
 
-.PHONY: test lint release e2e_test prep_local
+.PHONY: test lint build_release build_release_analytics e2e_test prep_local

README.md

@@ -14,11 +14,11 @@ Please see the [Python integration docs](https://posthog.com/docs/integrations/p
 
 ## Python Version Support
 
-| SDK Version | Python Versions Supported | Notes |
-|-------------|---------------------------|-------|
-| 7.3.1+      | 3.10, 3.11, 3.12, 3.13, 3.14 | Added Python 3.14 support |
-| 7.0.0 - 7.0.1 | 3.10, 3.11, 3.12, 3.13  | Dropped Python 3.9 support |
-| 4.0.1 - 6.x | 3.9, 3.10, 3.11, 3.12, 3.13 | Python 3.9+ required |
+| SDK Version   | Python Versions Supported    | Notes                      |
+| ------------- | ---------------------------- | -------------------------- |
+| 7.3.1+        | 3.10, 3.11, 3.12, 3.13, 3.14 | Added Python 3.14 support  |
+| 7.0.0 - 7.0.1 | 3.10, 3.11, 3.12, 3.13       | Dropped Python 3.9 support |
+| 4.0.1 - 6.x   | 3.9, 3.10, 3.11, 3.12, 3.13  | Python 3.9+ required       |
 
 ## Development
 
@@ -27,13 +27,13 @@ Please see the [Python integration docs](https://posthog.com/docs/integrations/p
 We recommend using [uv](https://docs.astral.sh/uv/). It's super fast.
 
 1. Run `uv venv env` (creates virtual environment called "env")
-    * or `python3 -m venv env`
+   - or `python3 -m venv env`
 2. Run `source env/bin/activate` (activates the virtual environment)
 3. Run `uv sync --extra dev --extra test` (installs the package in develop mode, along with test dependencies)
-    * or `pip install -e ".[dev,test]"`
+   - or `pip install -e ".[dev,test]"`
 4. you have to run `pre-commit install` to have auto linting pre commit
 5. Run `make test`
-  1. To run a specific test do `pytest -k test_no_api_key`
+6. To run a specific test do `pytest -k test_no_api_key`
 
 ## PostHog recommends `uv` so...
 
@@ -51,16 +51,10 @@ make test
 
 Assuming you have a [local version of PostHog](https://posthog.com/docs/developing-locally) running, you can run `python3 example.py` to see the library in action.
 
-### Releasing Versions
-
-Updates are released automatically using GitHub Actions when `version.py` is updated on `master`. After bumping `version.py` in `master` and adding to `CHANGELOG.md`, the [release workflow](https://github.com/PostHog/posthog-python/blob/master/.github/workflows/release.yaml) will automatically trigger and deploy the new version.
-
-If you need to check the latest runs or manually trigger a release, you can go to [our release workflow's page](https://github.com/PostHog/posthog-python/actions/workflows/release.yaml) and dispatch it manually, using workflow from `master`.
-
-
 ### Testing changes locally with the PostHog app
 
 You can run `make prep_local`, and it'll create a new folder alongside the SDK repo one called `posthog-python-local`, which you can then import into the posthog project by changing pyproject.toml to look like this:
+
 ```toml
 dependencies = [
     ...
@@ -71,4 +65,16 @@ dependencies = [
 [tools.uv.sources]
 posthoganalytics = { path = "../posthog-python-local" }
 ```
+
 This'll let you build and test SDK changes fully locally, incorporating them into your local posthog app stack. It mainly takes care of the `posthog -> posthoganalytics` module renaming. You'll need to re-run `make prep_local` each time you make a change, and re-run `uv sync --active` in the posthog app project.
+
+## Releasing
+
+This repository uses [Sampo](https://github.com/bruits/sampo) for versioning, changelogs, and publishing to crates.io.
+
+1. When making changes, include a changeset: `sampo add`
+2. Create a PR with your changes and the changeset file
+3. Add the `release` label and merge to `main`
+4. Approve the release in Slack when prompted — this triggers version bump, crates.io publish, git tag, and GitHub Release
+
+You can also trigger a release manually via the workflow's `workflow_dispatch` trigger (still requires pending changesets).

README_ANALYTICS.md

@@ -0,0 +1,11 @@
+# posthoganalytics
+
+> **Do not use this package.** Use [`posthog`](https://pypi.org/project/posthog/) instead.
+
+```bash
+pip install posthog
+```
+
+This package exists solely for internal use by [posthog/posthog](https://github.com/posthog/posthog) to avoid import conflicts with the local `posthog` package in that repository. It is an automatically generated mirror of `posthog` — same code, same versions, just published under a different name.
+
+If you are not working on the PostHog main repository, you should never need this package. All documentation, issues, and development happen in [`posthog-python`](https://github.com/posthog/posthog-python).

posthog/ai/__init__.py

@@ -0,0 +1,3 @@
+from posthog.ai.prompts import Prompts
+
+__all__ = ["Prompts"]

posthog/ai/anthropic/anthropic_converter.py

@@ -17,6 +17,7 @@ from posthog.ai.types import (
     TokenUsage,
     ToolInProgress,
 )
+from posthog.ai.utils import serialize_raw_usage
 
 
 def format_anthropic_response(response: Any) -> List[FormattedMessage]:
@@ -221,6 +222,12 @@ def extract_anthropic_usage_from_response(response: Any) -> TokenUsage:
     if web_search_count > 0:
         result["web_search_count"] = web_search_count
 
+    # Capture raw usage metadata for backend processing
+    # Serialize to dict here in the converter (not in utils)
+    serialized = serialize_raw_usage(response.usage)
+    if serialized:
+        result["raw_usage"] = serialized
+
     return result
 
 
@@ -247,6 +254,11 @@ def extract_anthropic_usage_from_event(event: Any) -> TokenUsage:
             usage["cache_read_input_tokens"] = getattr(
                 event.message.usage, "cache_read_input_tokens", 0
             )
+            # Capture raw usage metadata for backend processing
+            # Serialize to dict here in the converter (not in utils)
+            serialized = serialize_raw_usage(event.message.usage)
+            if serialized:
+                usage["raw_usage"] = serialized
 
     # Handle usage stats from message_delta event
     if hasattr(event, "usage") and event.usage:
@@ -262,6 +274,12 @@ def extract_anthropic_usage_from_event(event: Any) -> TokenUsage:
                 if web_search_count > 0:
                     usage["web_search_count"] = web_search_count
 
+        # Capture raw usage metadata for backend processing
+        # Serialize to dict here in the converter (not in utils)
+        serialized = serialize_raw_usage(event.usage)
+        if serialized:
+            usage["raw_usage"] = serialized
+
     return usage
 
 

posthog/ai/gemini/gemini_converter.py

@@ -12,6 +12,7 @@ from posthog.ai.types import (
     FormattedMessage,
     TokenUsage,
 )
+from posthog.ai.utils import serialize_raw_usage
 
 
 class GeminiPart(TypedDict, total=False):
@@ -487,6 +488,12 @@ def _extract_usage_from_metadata(metadata: Any) -> TokenUsage:
         if reasoning_tokens and reasoning_tokens > 0:
             usage["reasoning_tokens"] = reasoning_tokens
 
+    # Capture raw usage metadata for backend processing
+    # Serialize to dict here in the converter (not in utils)
+    serialized = serialize_raw_usage(metadata)
+    if serialized:
+        usage["raw_usage"] = serialized
+
     return usage
 
 

posthog/ai/openai/openai_converter.py

@@ -16,6 +16,7 @@ from posthog.ai.types import (
     FormattedTextContent,
     TokenUsage,
 )
+from posthog.ai.utils import serialize_raw_usage
 
 
 def format_openai_response(response: Any) -> List[FormattedMessage]:
@@ -429,6 +430,12 @@ def extract_openai_usage_from_response(response: Any) -> TokenUsage:
     if web_search_count > 0:
         result["web_search_count"] = web_search_count
 
+    # Capture raw usage metadata for backend processing
+    # Serialize to dict here in the converter (not in utils)
+    serialized = serialize_raw_usage(response.usage)
+    if serialized:
+        result["raw_usage"] = serialized
+
     return result
 
 
@@ -482,6 +489,12 @@ def extract_openai_usage_from_chunk(
                 chunk.usage.completion_tokens_details.reasoning_tokens
             )
 
+        # Capture raw usage metadata for backend processing
+        # Serialize to dict here in the converter (not in utils)
+        serialized = serialize_raw_usage(chunk.usage)
+        if serialized:
+            usage["raw_usage"] = serialized
+
     elif provider_type == "responses":
         # For Responses API, usage is only in chunk.response.usage for completed events
         if hasattr(chunk, "type") and chunk.type == "response.completed":
@@ -516,6 +529,12 @@ def extract_openai_usage_from_chunk(
                     if web_search_count > 0:
                         usage["web_search_count"] = web_search_count
 
+                # Capture raw usage metadata for backend processing
+                # Serialize to dict here in the converter (not in utils)
+                serialized = serialize_raw_usage(response_usage)
+                if serialized:
+                    usage["raw_usage"] = serialized
+
     return usage
 
 

posthog/ai/openai_agents/__init__.py

@@ -0,0 +1,76 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, Any, Callable, Dict, Optional, Union
+
+if TYPE_CHECKING:
+    from agents.tracing import Trace
+
+    from posthog.client import Client
+
+try:
+    import agents  # noqa: F401
+except ImportError:
+    raise ModuleNotFoundError(
+        "Please install the OpenAI Agents SDK to use this feature: 'pip install openai-agents'"
+    )
+
+from posthog.ai.openai_agents.processor import PostHogTracingProcessor
+
+__all__ = ["PostHogTracingProcessor", "instrument"]
+
+
+def instrument(
+    client: Optional[Client] = None,
+    distinct_id: Optional[Union[str, Callable[[Trace], Optional[str]]]] = None,
+    privacy_mode: bool = False,
+    groups: Optional[Dict[str, Any]] = None,
+    properties: Optional[Dict[str, Any]] = None,
+) -> PostHogTracingProcessor:
+    """
+    One-liner to instrument OpenAI Agents SDK with PostHog tracing.
+
+    This registers a PostHogTracingProcessor with the OpenAI Agents SDK,
+    automatically capturing traces, spans, and LLM generations.
+
+    Args:
+        client: Optional PostHog client instance. If not provided, uses the default client.
+        distinct_id: Optional distinct ID to associate with all traces.
+            Can also be a callable that takes a trace and returns a distinct ID.
+        privacy_mode: If True, redacts input/output content from events.
+        groups: Optional PostHog groups to associate with events.
+        properties: Optional additional properties to include with all events.
+
+    Returns:
+        PostHogTracingProcessor: The registered processor instance.
+
+    Example:
+        ```python
+        from posthog.ai.openai_agents import instrument
+
+        # Simple setup
+        instrument(distinct_id="user@example.com")
+
+        # With custom properties
+        instrument(
+            distinct_id="user@example.com",
+            privacy_mode=True,
+            properties={"environment": "production"}
+        )
+
+        # Now run agents as normal - traces automatically sent to PostHog
+        from agents import Agent, Runner
+        agent = Agent(name="Assistant", instructions="You are helpful.")
+        result = Runner.run_sync(agent, "Hello!")
+        ```
+    """
+    from agents.tracing import add_trace_processor
+
+    processor = PostHogTracingProcessor(
+        client=client,
+        distinct_id=distinct_id,
+        privacy_mode=privacy_mode,
+        groups=groups,
+        properties=properties,
+    )
+    add_trace_processor(processor)
+    return processor

posthog/ai/openai_agents/processor.py

@@ -0,0 +1,863 @@
+import json
+import logging
+import time
+from datetime import datetime
+from typing import Any, Callable, Dict, Optional, Union
+
+from agents.tracing import Span, Trace
+from agents.tracing.processor_interface import TracingProcessor
+from agents.tracing.span_data import (
+    AgentSpanData,
+    CustomSpanData,
+    FunctionSpanData,
+    GenerationSpanData,
+    GuardrailSpanData,
+    HandoffSpanData,
+    MCPListToolsSpanData,
+    ResponseSpanData,
+    SpeechGroupSpanData,
+    SpeechSpanData,
+    TranscriptionSpanData,
+)
+
+from posthog import setup
+from posthog.client import Client
+
+log = logging.getLogger("posthog")
+
+
+def _ensure_serializable(obj: Any) -> Any:
+    """Ensure an object is JSON-serializable, converting to str as fallback.
+
+    Returns the original object if it's already serializable (dict, list, str,
+    int, etc.), or str(obj) for non-serializable types so that downstream
+    json.dumps() calls won't fail.
+    """
+    if obj is None:
+        return None
+    try:
+        json.dumps(obj)
+        return obj
+    except (TypeError, ValueError):
+        return str(obj)
+
+
+def _parse_iso_timestamp(iso_str: Optional[str]) -> Optional[float]:
+    """Parse ISO timestamp to Unix timestamp."""
+    if not iso_str:
+        return None
+    try:
+        dt = datetime.fromisoformat(iso_str.replace("Z", "+00:00"))
+        return dt.timestamp()
+    except (ValueError, AttributeError):
+        return None
+
+
+class PostHogTracingProcessor(TracingProcessor):
+    """
+    A tracing processor that sends OpenAI Agents SDK traces to PostHog.
+
+    This processor implements the TracingProcessor interface from the OpenAI Agents SDK
+    and maps agent traces, spans, and generations to PostHog's LLM analytics events.
+
+    Example:
+        ```python
+        from agents import Agent, Runner
+        from agents.tracing import add_trace_processor
+        from posthog.ai.openai_agents import PostHogTracingProcessor
+
+        # Create and register the processor
+        processor = PostHogTracingProcessor(
+            distinct_id="user@example.com",
+            privacy_mode=False,
+        )
+        add_trace_processor(processor)
+
+        # Run agents as normal - traces automatically sent to PostHog
+        agent = Agent(name="Assistant", instructions="You are helpful.")
+        result = Runner.run_sync(agent, "Hello!")
+        ```
+    """
+
+    def __init__(
+        self,
+        client: Optional[Client] = None,
+        distinct_id: Optional[Union[str, Callable[[Trace], Optional[str]]]] = None,
+        privacy_mode: bool = False,
+        groups: Optional[Dict[str, Any]] = None,
+        properties: Optional[Dict[str, Any]] = None,
+    ):
+        """
+        Initialize the PostHog tracing processor.
+
+        Args:
+            client: Optional PostHog client instance. If not provided, uses the default client.
+            distinct_id: Either a string distinct ID or a callable that takes a Trace
+                and returns a distinct ID. If not provided, uses the trace_id.
+            privacy_mode: If True, redacts input/output content from events.
+            groups: Optional PostHog groups to associate with all events.
+            properties: Optional additional properties to include with all events.
+        """
+        self._client = client or setup()
+        self._distinct_id = distinct_id
+        self._privacy_mode = privacy_mode
+        self._groups = groups or {}
+        self._properties = properties or {}
+
+        # Track span start times for latency calculation
+        self._span_start_times: Dict[str, float] = {}
+
+        # Track trace metadata for associating with spans
+        self._trace_metadata: Dict[str, Dict[str, Any]] = {}
+
+        # Max entries to prevent unbounded growth if on_span_end/on_trace_end
+        # is never called (e.g., due to an exception in the Agents SDK).
+        self._max_tracked_entries = 10000
+
+    def _get_distinct_id(self, trace: Optional[Trace]) -> Optional[str]:
+        """Resolve the distinct ID for a trace.
+
+        Returns the user-provided distinct ID (string or callable result),
+        or None if no user-provided ID is available. Callers should treat
+        None as a signal to use a fallback ID in personless mode.
+        """
+        if callable(self._distinct_id):
+            if trace:
+                result = self._distinct_id(trace)
+                if result:
+                    return str(result)
+            return None
+        elif self._distinct_id:
+            return str(self._distinct_id)
+        return None
+
+    def _with_privacy_mode(self, value: Any) -> Any:
+        """Apply privacy mode redaction if enabled."""
+        if self._privacy_mode or (
+            hasattr(self._client, "privacy_mode") and self._client.privacy_mode
+        ):
+            return None
+        return value
+
+    def _evict_stale_entries(self) -> None:
+        """Evict oldest entries if dicts exceed max size to prevent unbounded growth."""
+        if len(self._span_start_times) > self._max_tracked_entries:
+            # Remove oldest entries by start time
+            sorted_spans = sorted(self._span_start_times.items(), key=lambda x: x[1])
+            for span_id, _ in sorted_spans[: len(sorted_spans) // 2]:
+                del self._span_start_times[span_id]
+            log.debug(
+                "Evicted stale span start times (exceeded %d entries)",
+                self._max_tracked_entries,
+            )
+
+        if len(self._trace_metadata) > self._max_tracked_entries:
+            # Remove half the entries (oldest inserted via dict ordering in Python 3.7+)
+            keys = list(self._trace_metadata.keys())
+            for key in keys[: len(keys) // 2]:
+                del self._trace_metadata[key]
+            log.debug(
+                "Evicted stale trace metadata (exceeded %d entries)",
+                self._max_tracked_entries,
+            )
+
+    def _get_group_id(self, trace_id: str) -> Optional[str]:
+        """Get the group_id for a trace from stored metadata."""
+        if trace_id in self._trace_metadata:
+            return self._trace_metadata[trace_id].get("group_id")
+        return None
+
+    def _capture_event(
+        self,
+        event: str,
+        properties: Dict[str, Any],
+        distinct_id: Optional[str] = None,
+    ) -> None:
+        """Capture an event to PostHog with error handling.
+
+        Args:
+            distinct_id: The resolved distinct ID. When the user didn't provide
+                one, callers should pass ``user_distinct_id or fallback_id``
+                (matching the langchain/openai pattern) and separately set
+                ``$process_person_profile`` in properties.
+        """
+        try:
+            if not hasattr(self._client, "capture") or not callable(
+                self._client.capture
+            ):
+                return
+
+            final_properties = {
+                **properties,
+                **self._properties,
+            }
+
+            self._client.capture(
+                distinct_id=distinct_id or "unknown",
+                event=event,
+                properties=final_properties,
+                groups=self._groups,
+            )
+        except Exception as e:
+            log.debug(f"Failed to capture PostHog event: {e}")
+
+    def on_trace_start(self, trace: Trace) -> None:
+        """Called when a new trace begins. Stores metadata for spans; the $ai_trace event is emitted in on_trace_end."""
+        try:
+            self._evict_stale_entries()
+            trace_id = trace.trace_id
+            trace_name = trace.name
+            group_id = getattr(trace, "group_id", None)
+            metadata = getattr(trace, "metadata", None)
+
+            distinct_id = self._get_distinct_id(trace)
+
+            # Store trace metadata for later (used by spans and on_trace_end)
+            self._trace_metadata[trace_id] = {
+                "name": trace_name,
+                "group_id": group_id,
+                "metadata": metadata,
+                "distinct_id": distinct_id,
+                "start_time": time.time(),
+            }
+        except Exception as e:
+            log.debug(f"Error in on_trace_start: {e}")
+
+    def on_trace_end(self, trace: Trace) -> None:
+        """Called when a trace completes. Emits the $ai_trace event with full metadata."""
+        try:
+            trace_id = trace.trace_id
+
+            # Pop stored metadata (also cleans up)
+            trace_info = self._trace_metadata.pop(trace_id, {})
+            trace_name = trace_info.get("name") or trace.name
+            group_id = trace_info.get("group_id") or getattr(trace, "group_id", None)
+            metadata = trace_info.get("metadata") or getattr(trace, "metadata", None)
+            distinct_id = trace_info.get("distinct_id") or self._get_distinct_id(trace)
+
+            # Calculate trace-level latency
+            start_time = trace_info.get("start_time")
+            latency = (time.time() - start_time) if start_time else None
+
+            properties = {
+                "$ai_trace_id": trace_id,
+                "$ai_trace_name": trace_name,
+                "$ai_provider": "openai",
+                "$ai_framework": "openai-agents",
+            }
+
+            if latency is not None:
+                properties["$ai_latency"] = latency
+
+            # Include group_id for linking related traces (e.g., conversation threads)
+            if group_id:
+                properties["$ai_group_id"] = group_id
+
+            # Include trace metadata if present
+            if metadata:
+                properties["$ai_trace_metadata"] = _ensure_serializable(metadata)
+
+            if distinct_id is None:
+                properties["$process_person_profile"] = False
+
+            self._capture_event(
+                event="$ai_trace",
+                distinct_id=distinct_id or trace_id,
+                properties=properties,
+            )
+        except Exception as e:
+            log.debug(f"Error in on_trace_end: {e}")
+
+    def on_span_start(self, span: Span[Any]) -> None:
+        """Called when a new span begins."""
+        try:
+            self._evict_stale_entries()
+            span_id = span.span_id
+            self._span_start_times[span_id] = time.time()
+        except Exception as e:
+            log.debug(f"Error in on_span_start: {e}")
+
+    def on_span_end(self, span: Span[Any]) -> None:
+        """Called when a span completes."""
+        try:
+            span_id = span.span_id
+            trace_id = span.trace_id
+            parent_id = span.parent_id
+            span_data = span.span_data
+
+            # Calculate latency
+            start_time = self._span_start_times.pop(span_id, None)
+            if start_time:
+                latency = time.time() - start_time
+            else:
+                # Fall back to parsing timestamps
+                started = _parse_iso_timestamp(span.started_at)
+                ended = _parse_iso_timestamp(span.ended_at)
+                latency = (ended - started) if (started and ended) else 0
+
+            # Get user-provided distinct ID from trace metadata (resolved at trace start).
+            # None means no user-provided ID — use trace_id as fallback in personless mode,
+            # matching the langchain/openai pattern: `distinct_id or trace_id`.
+            trace_info = self._trace_metadata.get(trace_id, {})
+            distinct_id = trace_info.get("distinct_id") or self._get_distinct_id(None)
+
+            # Get group_id from trace metadata for linking
+            group_id = self._get_group_id(trace_id)
+
+            # Get error info if present
+            error_info = span.error
+            error_properties = {}
+            if error_info:
+                if isinstance(error_info, dict):
+                    error_message = error_info.get("message", str(error_info))
+                    error_type_raw = error_info.get("type", "")
+                else:
+                    error_message = str(error_info)
+                    error_type_raw = ""
+
+                # Categorize error type for cross-provider filtering/alerting
+                error_type = "unknown"
+                if (
+                    "ModelBehaviorError" in error_type_raw
+                    or "ModelBehaviorError" in error_message
+                ):
+                    error_type = "model_behavior_error"
+                elif "UserError" in error_type_raw or "UserError" in error_message:
+                    error_type = "user_error"
+                elif (
+                    "InputGuardrailTripwireTriggered" in error_type_raw
+                    or "InputGuardrailTripwireTriggered" in error_message
+                ):
+                    error_type = "input_guardrail_triggered"
+                elif (
+                    "OutputGuardrailTripwireTriggered" in error_type_raw
+                    or "OutputGuardrailTripwireTriggered" in error_message
+                ):
+                    error_type = "output_guardrail_triggered"
+                elif (
+                    "MaxTurnsExceeded" in error_type_raw
+                    or "MaxTurnsExceeded" in error_message
+                ):
+                    error_type = "max_turns_exceeded"
+
+                error_properties = {
+                    "$ai_is_error": True,
+                    "$ai_error": error_message,
+                    "$ai_error_type": error_type,
+                }
+
+            # Personless mode: no user-provided distinct_id, fallback to trace_id
+            if distinct_id is None:
+                error_properties["$process_person_profile"] = False
+                distinct_id = trace_id
+
+            # Dispatch based on span data type
+            if isinstance(span_data, GenerationSpanData):
+                self._handle_generation_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            elif isinstance(span_data, FunctionSpanData):
+                self._handle_function_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            elif isinstance(span_data, AgentSpanData):
+                self._handle_agent_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            elif isinstance(span_data, HandoffSpanData):
+                self._handle_handoff_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            elif isinstance(span_data, GuardrailSpanData):
+                self._handle_guardrail_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            elif isinstance(span_data, ResponseSpanData):
+                self._handle_response_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            elif isinstance(span_data, CustomSpanData):
+                self._handle_custom_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            elif isinstance(
+                span_data, (TranscriptionSpanData, SpeechSpanData, SpeechGroupSpanData)
+            ):
+                self._handle_audio_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            elif isinstance(span_data, MCPListToolsSpanData):
+                self._handle_mcp_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+            else:
+                # Unknown span type - capture as generic span
+                self._handle_generic_span(
+                    span_data,
+                    trace_id,
+                    span_id,
+                    parent_id,
+                    latency,
+                    distinct_id,
+                    group_id,
+                    error_properties,
+                )
+
+        except Exception as e:
+            log.debug(f"Error in on_span_end: {e}")
+
+    def _base_properties(
+        self,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> Dict[str, Any]:
+        """Build the base properties dict shared by all span handlers."""
+        properties = {
+            "$ai_trace_id": trace_id,
+            "$ai_span_id": span_id,
+            "$ai_parent_id": parent_id,
+            "$ai_provider": "openai",
+            "$ai_framework": "openai-agents",
+            "$ai_latency": latency,
+            **error_properties,
+        }
+        if group_id:
+            properties["$ai_group_id"] = group_id
+        return properties
+
+    def _handle_generation_span(
+        self,
+        span_data: GenerationSpanData,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle LLM generation spans - maps to $ai_generation event."""
+        # Extract token usage
+        usage = span_data.usage or {}
+        input_tokens = usage.get("input_tokens") or usage.get("prompt_tokens") or 0
+        output_tokens = (
+            usage.get("output_tokens") or usage.get("completion_tokens") or 0
+        )
+
+        # Extract model config parameters
+        model_config = span_data.model_config or {}
+        model_params = {}
+        for param in [
+            "temperature",
+            "max_tokens",
+            "top_p",
+            "frequency_penalty",
+            "presence_penalty",
+        ]:
+            if param in model_config:
+                model_params[param] = model_config[param]
+
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_model": span_data.model,
+            "$ai_model_parameters": model_params if model_params else None,
+            "$ai_input": self._with_privacy_mode(_ensure_serializable(span_data.input)),
+            "$ai_output_choices": self._with_privacy_mode(
+                _ensure_serializable(span_data.output)
+            ),
+            "$ai_input_tokens": input_tokens,
+            "$ai_output_tokens": output_tokens,
+            "$ai_total_tokens": (input_tokens or 0) + (output_tokens or 0),
+        }
+
+        # Add optional token fields if present
+        if usage.get("reasoning_tokens"):
+            properties["$ai_reasoning_tokens"] = usage["reasoning_tokens"]
+        if usage.get("cache_read_input_tokens"):
+            properties["$ai_cache_read_input_tokens"] = usage["cache_read_input_tokens"]
+        if usage.get("cache_creation_input_tokens"):
+            properties["$ai_cache_creation_input_tokens"] = usage[
+                "cache_creation_input_tokens"
+            ]
+
+        self._capture_event("$ai_generation", properties, distinct_id)
+
+    def _handle_function_span(
+        self,
+        span_data: FunctionSpanData,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle function/tool call spans - maps to $ai_span event."""
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_span_name": span_data.name,
+            "$ai_span_type": "tool",
+            "$ai_input_state": self._with_privacy_mode(
+                _ensure_serializable(span_data.input)
+            ),
+            "$ai_output_state": self._with_privacy_mode(
+                _ensure_serializable(span_data.output)
+            ),
+        }
+
+        if span_data.mcp_data:
+            properties["$ai_mcp_data"] = _ensure_serializable(span_data.mcp_data)
+
+        self._capture_event("$ai_span", properties, distinct_id)
+
+    def _handle_agent_span(
+        self,
+        span_data: AgentSpanData,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle agent execution spans - maps to $ai_span event."""
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_span_name": span_data.name,
+            "$ai_span_type": "agent",
+        }
+
+        if span_data.handoffs:
+            properties["$ai_agent_handoffs"] = span_data.handoffs
+        if span_data.tools:
+            properties["$ai_agent_tools"] = span_data.tools
+        if span_data.output_type:
+            properties["$ai_agent_output_type"] = span_data.output_type
+
+        self._capture_event("$ai_span", properties, distinct_id)
+
+    def _handle_handoff_span(
+        self,
+        span_data: HandoffSpanData,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle agent handoff spans - maps to $ai_span event."""
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_span_name": f"{span_data.from_agent} -> {span_data.to_agent}",
+            "$ai_span_type": "handoff",
+            "$ai_handoff_from_agent": span_data.from_agent,
+            "$ai_handoff_to_agent": span_data.to_agent,
+        }
+
+        self._capture_event("$ai_span", properties, distinct_id)
+
+    def _handle_guardrail_span(
+        self,
+        span_data: GuardrailSpanData,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle guardrail execution spans - maps to $ai_span event."""
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_span_name": span_data.name,
+            "$ai_span_type": "guardrail",
+            "$ai_guardrail_triggered": span_data.triggered,
+        }
+
+        self._capture_event("$ai_span", properties, distinct_id)
+
+    def _handle_response_span(
+        self,
+        span_data: ResponseSpanData,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle OpenAI Response API spans - maps to $ai_generation event."""
+        response = span_data.response
+        response_id = response.id if response else None
+
+        # Try to extract usage from response
+        usage = getattr(response, "usage", None) if response else None
+        input_tokens = 0
+        output_tokens = 0
+        if usage:
+            input_tokens = getattr(usage, "input_tokens", 0) or 0
+            output_tokens = getattr(usage, "output_tokens", 0) or 0
+
+        # Try to extract model from response
+        model = getattr(response, "model", None) if response else None
+
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_model": model,
+            "$ai_response_id": response_id,
+            "$ai_input": self._with_privacy_mode(_ensure_serializable(span_data.input)),
+            "$ai_input_tokens": input_tokens,
+            "$ai_output_tokens": output_tokens,
+            "$ai_total_tokens": input_tokens + output_tokens,
+        }
+
+        # Extract output content from response
+        if response:
+            output_items = getattr(response, "output", None)
+            if output_items:
+                properties["$ai_output_choices"] = self._with_privacy_mode(
+                    _ensure_serializable(output_items)
+                )
+
+        self._capture_event("$ai_generation", properties, distinct_id)
+
+    def _handle_custom_span(
+        self,
+        span_data: CustomSpanData,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle custom user-defined spans - maps to $ai_span event."""
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_span_name": span_data.name,
+            "$ai_span_type": "custom",
+            "$ai_custom_data": self._with_privacy_mode(
+                _ensure_serializable(span_data.data)
+            ),
+        }
+
+        self._capture_event("$ai_span", properties, distinct_id)
+
+    def _handle_audio_span(
+        self,
+        span_data: Union[TranscriptionSpanData, SpeechSpanData, SpeechGroupSpanData],
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle audio-related spans (transcription, speech) - maps to $ai_span event."""
+        span_type = span_data.type  # "transcription", "speech", or "speech_group"
+
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_span_name": span_type,
+            "$ai_span_type": span_type,
+        }
+
+        # Add model info if available
+        if hasattr(span_data, "model") and span_data.model:
+            properties["$ai_model"] = span_data.model
+
+        # Add model config if available (pass-through property)
+        if hasattr(span_data, "model_config") and span_data.model_config:
+            properties["model_config"] = _ensure_serializable(span_data.model_config)
+
+        # Add time to first audio byte for speech spans (pass-through property)
+        if hasattr(span_data, "first_content_at") and span_data.first_content_at:
+            properties["first_content_at"] = span_data.first_content_at
+
+        # Add audio format info (pass-through properties)
+        if hasattr(span_data, "input_format"):
+            properties["audio_input_format"] = span_data.input_format
+        if hasattr(span_data, "output_format"):
+            properties["audio_output_format"] = span_data.output_format
+
+        # Add text input for TTS
+        if (
+            hasattr(span_data, "input")
+            and span_data.input
+            and isinstance(span_data.input, str)
+        ):
+            properties["$ai_input"] = self._with_privacy_mode(span_data.input)
+
+        # Don't include audio data (base64) - just metadata
+        if hasattr(span_data, "output") and isinstance(span_data.output, str):
+            # For transcription, output is the text
+            properties["$ai_output_state"] = self._with_privacy_mode(span_data.output)
+
+        self._capture_event("$ai_span", properties, distinct_id)
+
+    def _handle_mcp_span(
+        self,
+        span_data: MCPListToolsSpanData,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle MCP (Model Context Protocol) spans - maps to $ai_span event."""
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_span_name": f"mcp:{span_data.server}",
+            "$ai_span_type": "mcp_tools",
+            "$ai_mcp_server": span_data.server,
+            "$ai_mcp_tools": span_data.result,
+        }
+
+        self._capture_event("$ai_span", properties, distinct_id)
+
+    def _handle_generic_span(
+        self,
+        span_data: Any,
+        trace_id: str,
+        span_id: str,
+        parent_id: Optional[str],
+        latency: float,
+        distinct_id: str,
+        group_id: Optional[str],
+        error_properties: Dict[str, Any],
+    ) -> None:
+        """Handle unknown span types - maps to $ai_span event."""
+        span_type = getattr(span_data, "type", "unknown")
+
+        properties = {
+            **self._base_properties(
+                trace_id, span_id, parent_id, latency, group_id, error_properties
+            ),
+            "$ai_span_name": span_type,
+            "$ai_span_type": span_type,
+        }
+
+        # Try to export span data
+        if hasattr(span_data, "export"):
+            try:
+                exported = span_data.export()
+                properties["$ai_span_data"] = _ensure_serializable(exported)
+            except Exception:
+                pass
+
+        self._capture_event("$ai_span", properties, distinct_id)
+
+    def shutdown(self) -> None:
+        """Clean up resources when the application stops."""
+        try:
+            self._span_start_times.clear()
+            self._trace_metadata.clear()
+
+            # Flush the PostHog client if possible
+            if hasattr(self._client, "flush") and callable(self._client.flush):
+                self._client.flush()
+        except Exception as e:
+            log.debug(f"Error in shutdown: {e}")
+
+    def force_flush(self) -> None:
+        """Force immediate processing of any queued events."""
+        try:
+            if hasattr(self._client, "flush") and callable(self._client.flush):
+                self._client.flush()
+        except Exception as e:
+            log.debug(f"Error in force_flush: {e}")

posthog/ai/prompts.py

@@ -0,0 +1,286 @@
+"""
+Prompt management for PostHog AI SDK.
+
+Fetch and compile LLM prompts from PostHog with caching and fallback support.
+"""
+
+import logging
+import re
+import time
+import urllib.parse
+from typing import Any, Dict, Optional, Union
+
+from posthog.request import USER_AGENT, _get_session
+from posthog.utils import remove_trailing_slash
+
+log = logging.getLogger("posthog")
+
+APP_ENDPOINT = "https://us.posthog.com"
+DEFAULT_CACHE_TTL_SECONDS = 300  # 5 minutes
+
+PromptVariables = Dict[str, Union[str, int, float, bool]]
+
+
+class CachedPrompt:
+    """Cached prompt with metadata."""
+
+    def __init__(self, prompt: str, fetched_at: float):
+        self.prompt = prompt
+        self.fetched_at = fetched_at
+
+
+def _is_prompt_api_response(data: Any) -> bool:
+    """Check if the response is a valid prompt API response."""
+    return (
+        isinstance(data, dict)
+        and "prompt" in data
+        and isinstance(data.get("prompt"), str)
+    )
+
+
+class Prompts:
+    """
+    Fetch and compile LLM prompts from PostHog.
+
+    Can be initialized with a PostHog client or with direct options.
+
+    Examples:
+        ```python
+        from posthog import Posthog
+        from posthog.ai.prompts import Prompts
+
+        # With PostHog client
+        posthog = Posthog('phc_xxx', host='https://us.posthog.com', personal_api_key='phx_xxx')
+        prompts = Prompts(posthog)
+
+        # Or with direct options (no PostHog client needed)
+        prompts = Prompts(
+            personal_api_key='phx_xxx',
+            project_api_key='phc_xxx',
+            host='https://us.posthog.com',
+        )
+
+        # Fetch with caching and fallback
+        template = prompts.get('support-system-prompt', fallback='You are a helpful assistant.')
+
+        # Compile with variables
+        system_prompt = prompts.compile(template, {
+            'company': 'Acme Corp',
+            'tier': 'premium',
+        })
+        ```
+    """
+
+    def __init__(
+        self,
+        posthog: Optional[Any] = None,
+        *,
+        personal_api_key: Optional[str] = None,
+        project_api_key: Optional[str] = None,
+        host: Optional[str] = None,
+        default_cache_ttl_seconds: Optional[int] = None,
+    ):
+        """
+        Initialize Prompts.
+
+        Args:
+            posthog: PostHog client instance (optional if personal_api_key provided)
+            personal_api_key: Direct personal API key (optional if posthog provided)
+            project_api_key: Direct project API key (optional if posthog provided)
+            host: PostHog host (defaults to app endpoint)
+            default_cache_ttl_seconds: Default cache TTL (defaults to 300)
+        """
+        self._default_cache_ttl_seconds = (
+            default_cache_ttl_seconds or DEFAULT_CACHE_TTL_SECONDS
+        )
+        self._cache: Dict[str, CachedPrompt] = {}
+
+        if posthog is not None:
+            self._personal_api_key = getattr(posthog, "personal_api_key", None) or ""
+            self._project_api_key = getattr(posthog, "api_key", None) or ""
+            self._host = remove_trailing_slash(
+                getattr(posthog, "raw_host", None) or APP_ENDPOINT
+            )
+        else:
+            self._personal_api_key = personal_api_key or ""
+            self._project_api_key = project_api_key or ""
+            self._host = remove_trailing_slash(host or APP_ENDPOINT)
+
+    def get(
+        self,
+        name: str,
+        *,
+        cache_ttl_seconds: Optional[int] = None,
+        fallback: Optional[str] = None,
+    ) -> str:
+        """
+        Fetch a prompt by name from the PostHog API.
+
+        Caching behavior:
+        1. If cache is fresh, return cached value
+        2. If fetch fails and cache exists (stale), return stale cache with warning
+        3. If fetch fails and fallback provided, return fallback with warning
+        4. If fetch fails with no cache/fallback, raise exception
+
+        Args:
+            name: The name of the prompt to fetch
+            cache_ttl_seconds: Cache TTL in seconds (defaults to instance default)
+            fallback: Fallback prompt to use if fetch fails and no cache available
+
+        Returns:
+            The prompt string
+
+        Raises:
+            Exception: If the prompt cannot be fetched and no fallback is available
+        """
+        ttl = (
+            cache_ttl_seconds
+            if cache_ttl_seconds is not None
+            else self._default_cache_ttl_seconds
+        )
+
+        # Check cache first
+        cached = self._cache.get(name)
+        now = time.time()
+
+        if cached is not None:
+            is_fresh = (now - cached.fetched_at) < ttl
+
+            if is_fresh:
+                return cached.prompt
+
+        # Try to fetch from API
+        try:
+            prompt = self._fetch_prompt_from_api(name)
+            fetched_at = time.time()
+
+            # Update cache
+            self._cache[name] = CachedPrompt(prompt=prompt, fetched_at=fetched_at)
+
+            return prompt
+
+        except Exception as error:
+            # Fallback order:
+            # 1. Return stale cache (with warning)
+            if cached is not None:
+                log.warning(
+                    '[PostHog Prompts] Failed to fetch prompt "%s", using stale cache: %s',
+                    name,
+                    error,
+                )
+                return cached.prompt
+
+            # 2. Return fallback (with warning)
+            if fallback is not None:
+                log.warning(
+                    '[PostHog Prompts] Failed to fetch prompt "%s", using fallback: %s',
+                    name,
+                    error,
+                )
+                return fallback
+
+            # 3. Raise error
+            raise
+
+    def compile(self, prompt: str, variables: PromptVariables) -> str:
+        """
+        Replace {{variableName}} placeholders with values.
+
+        Unmatched variables are left unchanged.
+        Supports variable names with hyphens and dots (e.g., user-id, company.name).
+
+        Args:
+            prompt: The prompt template string
+            variables: Object containing variable values
+
+        Returns:
+            The compiled prompt string
+        """
+
+        def replace_variable(match: re.Match) -> str:
+            variable_name = match.group(1)
+
+            if variable_name in variables:
+                return str(variables[variable_name])
+
+            return match.group(0)
+
+        return re.sub(r"\{\{([\w.-]+)\}\}", replace_variable, prompt)
+
+    def clear_cache(self, name: Optional[str] = None) -> None:
+        """
+        Clear cached prompts.
+
+        Args:
+            name: Specific prompt to clear. If None, clears all cached prompts.
+        """
+        if name is not None:
+            self._cache.pop(name, None)
+        else:
+            self._cache.clear()
+
+    def _fetch_prompt_from_api(self, name: str) -> str:
+        """
+        Fetch prompt from PostHog API.
+
+        Endpoint: {host}/api/environments/@current/llm_prompts/name/{encoded_name}/?token={encoded_project_api_key}
+        Auth: Bearer {personal_api_key}
+
+        Args:
+            name: The name of the prompt to fetch
+
+        Returns:
+            The prompt string
+
+        Raises:
+            Exception: If the prompt cannot be fetched
+        """
+        if not self._personal_api_key:
+            raise Exception(
+                "[PostHog Prompts] personal_api_key is required to fetch prompts. "
+                "Please provide it when initializing the Prompts instance."
+            )
+        if not self._project_api_key:
+            raise Exception(
+                "[PostHog Prompts] project_api_key is required to fetch prompts. "
+                "Please provide it when initializing the Prompts instance."
+            )
+
+        encoded_name = urllib.parse.quote(name, safe="")
+        encoded_project_api_key = urllib.parse.quote(self._project_api_key, safe="")
+        url = f"{self._host}/api/environments/@current/llm_prompts/name/{encoded_name}/?token={encoded_project_api_key}"
+
+        headers = {
+            "Authorization": f"Bearer {self._personal_api_key}",
+            "User-Agent": USER_AGENT,
+        }
+
+        response = _get_session().get(url, headers=headers, timeout=10)
+
+        if not response.ok:
+            if response.status_code == 404:
+                raise Exception(f'[PostHog Prompts] Prompt "{name}" not found')
+
+            if response.status_code == 403:
+                raise Exception(
+                    f'[PostHog Prompts] Access denied for prompt "{name}". '
+                    "Check that your personal_api_key has the correct permissions and the LLM prompts feature is enabled."
+                )
+
+            raise Exception(
+                f'[PostHog Prompts] Failed to fetch prompt "{name}": HTTP {response.status_code}'
+            )
+
+        try:
+            data = response.json()
+        except Exception:
+            raise Exception(
+                f'[PostHog Prompts] Invalid response format for prompt "{name}"'
+            )
+
+        if not _is_prompt_api_response(data):
+            raise Exception(
+                f'[PostHog Prompts] Invalid response format for prompt "{name}"'
+            )
+
+        return data["prompt"]

posthog/ai/sanitization.py

@@ -83,6 +83,12 @@ def sanitize_openai_image(item: Any) -> Any:
     if not isinstance(item, dict):
         return item
 
+    if item.get("type") == "input_image" and isinstance(item.get("image_url"), str):
+        return {
+            **item,
+            "image_url": redact_base64_data_url(item["image_url"]),
+        }
+
     if (
         item.get("type") == "image_url"
         and isinstance(item.get("image_url"), dict)

posthog/ai/types.py

@@ -64,6 +64,7 @@ class TokenUsage(TypedDict, total=False):
     cache_creation_input_tokens: Optional[int]
     reasoning_tokens: Optional[int]
     web_search_count: Optional[int]
+    raw_usage: Optional[Any]  # Raw provider usage metadata for backend processing
 
 
 class ProviderResponse(TypedDict, total=False):

posthog/ai/utils.py

@@ -2,7 +2,7 @@ import time
 import uuid
 from typing import Any, Callable, Dict, List, Optional, cast
 
-from posthog import get_tags, identify_context, new_context, tag
+from posthog import get_tags, identify_context, new_context, tag, contexts
 from posthog.ai.sanitization import (
     sanitize_anthropic,
     sanitize_gemini,
@@ -13,6 +13,76 @@ from posthog.ai.types import FormattedMessage, StreamingEventData, TokenUsage
 from posthog.client import Client as PostHogClient
 
 
+_TOKEN_PROPERTY_KEYS = frozenset(
+    {
+        "$ai_input_tokens",
+        "$ai_output_tokens",
+        "$ai_cache_read_input_tokens",
+        "$ai_cache_creation_input_tokens",
+        "$ai_total_tokens",
+        "$ai_reasoning_tokens",
+    }
+)
+
+
+def _get_tokens_source(
+    sdk_tags: Dict[str, Any], posthog_properties: Optional[Dict[str, Any]]
+) -> str:
+    if posthog_properties and any(
+        key in posthog_properties for key in _TOKEN_PROPERTY_KEYS
+    ):
+        return "passthrough"
+    return "sdk"
+
+
+def serialize_raw_usage(raw_usage: Any) -> Optional[Dict[str, Any]]:
+    """
+    Convert raw provider usage objects to JSON-serializable dicts.
+
+    Handles Pydantic models (OpenAI/Anthropic) and protobuf-like objects (Gemini)
+    with a fallback chain to ensure we never pass unserializable objects to PostHog.
+
+    Args:
+        raw_usage: Raw usage object from provider SDK
+
+    Returns:
+        Plain dict or None if conversion fails
+    """
+    if raw_usage is None:
+        return None
+
+    # Already a dict
+    if isinstance(raw_usage, dict):
+        return raw_usage
+
+    # Try Pydantic model_dump() (OpenAI/Anthropic)
+    if hasattr(raw_usage, "model_dump") and callable(raw_usage.model_dump):
+        try:
+            return raw_usage.model_dump()
+        except Exception:
+            pass
+
+    # Try to_dict() (some protobuf objects)
+    if hasattr(raw_usage, "to_dict") and callable(raw_usage.to_dict):
+        try:
+            return raw_usage.to_dict()
+        except Exception:
+            pass
+
+    # Try __dict__ / vars() for simple objects
+    try:
+        return vars(raw_usage)
+    except Exception:
+        pass
+
+    # Last resort: convert to string representation
+    # This ensures we always return something rather than failing
+    try:
+        return {"_raw": str(raw_usage)}
+    except Exception:
+        return None
+
+
 def merge_usage_stats(
     target: TokenUsage, source: TokenUsage, mode: str = "incremental"
 ) -> None:
@@ -60,6 +130,17 @@ def merge_usage_stats(
             current = target.get("web_search_count") or 0
             target["web_search_count"] = max(current, source_web_search)
 
+        # Merge raw_usage to avoid losing data from earlier events
+        # For Anthropic streaming: message_start has input tokens, message_delta has output
+        # Note: raw_usage is already serialized by converters, so it's a dict
+        source_raw_usage = source.get("raw_usage")
+        if source_raw_usage is not None and isinstance(source_raw_usage, dict):
+            current_raw_value = target.get("raw_usage")
+            current_raw: Dict[str, Any] = (
+                current_raw_value if isinstance(current_raw_value, dict) else {}
+            )
+            target["raw_usage"] = {**current_raw, **source_raw_usage}
+
     elif mode == "cumulative":
         # Replace with latest values (already cumulative)
         if source.get("input_tokens") is not None:
@@ -76,6 +157,9 @@ def merge_usage_stats(
             target["reasoning_tokens"] = source["reasoning_tokens"]
         if source.get("web_search_count") is not None:
             target["web_search_count"] = source["web_search_count"]
+        # Note: raw_usage is already serialized by converters, so it's a dict
+        if source.get("raw_usage") is not None:
+            target["raw_usage"] = source["raw_usage"]
 
     else:
         raise ValueError(f"Invalid mode: {mode}. Must be 'incremental' or 'cumulative'")
@@ -282,6 +366,16 @@ def call_llm_and_track_usage(
             if posthog_trace_id is None:
                 posthog_trace_id = str(uuid.uuid4())
 
+            # Check if we have a real user distinct_id (from param or outer context)
+            has_person_distinct_id = (
+                posthog_distinct_id is not None
+                or contexts.get_context_distinct_id() is not None
+            )
+
+            if not has_person_distinct_id:
+                # Fall back to trace_id as distinct_id when no real user id is available.
+                identify_context(posthog_trace_id)
+
             if response and (
                 hasattr(response, "usage")
                 or (provider == "gemini" and hasattr(response, "usage_metadata"))
@@ -332,7 +426,12 @@ def call_llm_and_track_usage(
             if web_search_count is not None and web_search_count > 0:
                 tag("$ai_web_search_count", web_search_count)
 
-            if posthog_distinct_id is None:
+            raw_usage = usage.get("raw_usage")
+            if raw_usage is not None:
+                # Already serialized by converters
+                tag("$ai_usage", raw_usage)
+
+            if not has_person_distinct_id:
                 tag("$process_person_profile", False)
 
             # Process instructions for Responses API
@@ -346,14 +445,19 @@ def call_llm_and_track_usage(
 
             # send the event to posthog
             if hasattr(ph_client, "capture") and callable(ph_client.capture):
+                sdk_tags = get_tags()
+                merged_properties = {
+                    **sdk_tags,
+                    **(posthog_properties or {}),
+                    **(error_params or {}),
+                }
+                merged_properties["$ai_tokens_source"] = _get_tokens_source(
+                    sdk_tags, posthog_properties
+                )
                 ph_client.capture(
-                    distinct_id=posthog_distinct_id or posthog_trace_id,
+                    distinct_id=contexts.get_context_distinct_id(),
                     event="$ai_generation",
-                    properties={
-                        **get_tags(),
-                        **(posthog_properties or {}),
-                        **(error_params or {}),
-                    },
+                    properties=merged_properties,
                     groups=posthog_groups,
                 )
 
@@ -407,6 +511,16 @@ async def call_llm_and_track_usage_async(
             if posthog_trace_id is None:
                 posthog_trace_id = str(uuid.uuid4())
 
+            # Check if we have a real user distinct_id (from param or outer context)
+            has_person_distinct_id = (
+                posthog_distinct_id is not None
+                or contexts.get_context_distinct_id() is not None
+            )
+
+            if not has_person_distinct_id:
+                # Fall back to trace_id as distinct_id when no real user id is available.
+                identify_context(posthog_trace_id)
+
             if response and (
                 hasattr(response, "usage")
                 or (provider == "gemini" and hasattr(response, "usage_metadata"))
@@ -457,7 +571,12 @@ async def call_llm_and_track_usage_async(
             if web_search_count is not None and web_search_count > 0:
                 tag("$ai_web_search_count", web_search_count)
 
-            if posthog_distinct_id is None:
+            raw_usage = usage.get("raw_usage")
+            if raw_usage is not None:
+                # Already serialized by converters
+                tag("$ai_usage", raw_usage)
+
+            if not has_person_distinct_id:
                 tag("$process_person_profile", False)
 
             # Process instructions for Responses API
@@ -471,14 +590,19 @@ async def call_llm_and_track_usage_async(
 
             # send the event to posthog
             if hasattr(ph_client, "capture") and callable(ph_client.capture):
+                sdk_tags = get_tags()
+                merged_properties = {
+                    **sdk_tags,
+                    **(posthog_properties or {}),
+                    **(error_params or {}),
+                }
+                merged_properties["$ai_tokens_source"] = _get_tokens_source(
+                    sdk_tags, posthog_properties
+                )
                 ph_client.capture(
-                    distinct_id=posthog_distinct_id or posthog_trace_id,
+                    distinct_id=contexts.get_context_distinct_id(),
                     event="$ai_generation",
-                    properties={
-                        **get_tags(),
-                        **(posthog_properties or {}),
-                        **(error_params or {}),
-                    },
+                    properties=merged_properties,
                     groups=posthog_groups,
                 )
 
@@ -555,6 +679,15 @@ def capture_streaming_event(
         **(event_data.get("properties") or {}),
     }
 
+    # Determine token source: SDK-computed vs externally overridden
+    sdk_token_tags = {
+        "$ai_input_tokens": event_data["usage_stats"].get("input_tokens", 0),
+        "$ai_output_tokens": event_data["usage_stats"].get("output_tokens", 0),
+    }
+    event_properties["$ai_tokens_source"] = _get_tokens_source(
+        sdk_token_tags, event_data.get("properties")
+    )
+
     # Extract and add tools based on provider
     available_tools = extract_available_tool_calls(
         event_data["provider"],
@@ -594,6 +727,12 @@ def capture_streaming_event(
     ):
         event_properties["$ai_web_search_count"] = web_search_count
 
+    # Add raw usage metadata if present (all providers)
+    raw_usage = event_data["usage_stats"].get("raw_usage")
+    if raw_usage is not None:
+        # Already serialized by converters
+        event_properties["$ai_usage"] = raw_usage
+
     # Handle provider-specific fields
     if (
         event_data["provider"] == "openai"

posthog/client.py

@@ -38,6 +38,7 @@ from posthog.feature_flags import (
     InconclusiveMatchError,
     RequiresServerEvaluation,
     match_feature_flag_properties,
+    resolve_bucketing_value,
 )
 from posthog.flag_definition_cache import (
     FlagDefinitionCacheData,
@@ -1335,6 +1336,13 @@ class Client(object):
                 self.log.error(
                     "[FEATURE FLAGS] Error loading feature flags: To use feature flags, please set a valid personal_api_key. More information: https://posthog.com/docs/api/overview"
                 )
+                self.feature_flags = []
+                self.group_type_mapping = {}
+                self.cohorts = {}
+
+                if self.flag_cache:
+                    self.flag_cache.clear()
+
                 if self.debug:
                     raise APIError(
                         status=401,
@@ -1411,6 +1419,7 @@ class Client(object):
         person_properties=None,
         group_properties=None,
         warn_on_unknown_groups=True,
+        device_id=None,
     ) -> FlagValue:
         groups = groups or {}
         person_properties = person_properties or {}
@@ -1451,22 +1460,35 @@ class Client(object):
                     )
                 return False
 
+            if group_name not in group_properties:
+                raise InconclusiveMatchError(
+                    f"Flag has no group properties for group '{group_name}'"
+                )
             focused_group_properties = group_properties[group_name]
+            group_key = groups[group_name]
             return match_feature_flag_properties(
                 feature_flag,
-                groups[group_name],
+                group_key,
                 focused_group_properties,
-                self.feature_flags_by_key,
-                evaluation_cache,
+                cohort_properties=self.cohorts,
+                flags_by_key=self.feature_flags_by_key,
+                evaluation_cache=evaluation_cache,
+                device_id=device_id,
+                bucketing_value=group_key,
             )
         else:
+            bucketing_value = resolve_bucketing_value(
+                feature_flag, distinct_id, device_id
+            )
             return match_feature_flag_properties(
                 feature_flag,
                 distinct_id,
                 person_properties,
-                self.cohorts,
-                self.feature_flags_by_key,
-                evaluation_cache,
+                cohort_properties=self.cohorts,
+                flags_by_key=self.feature_flags_by_key,
+                evaluation_cache=evaluation_cache,
+                device_id=device_id,
+                bucketing_value=bucketing_value,
             )
 
     def feature_enabled(
@@ -1573,8 +1595,12 @@ class Client(object):
         evaluated_at = None
         feature_flag_error: Optional[str] = None
 
+        # Resolve device_id from context if not provided
+        if device_id is None:
+            device_id = get_context_device_id()
+
         flag_value = self._locally_evaluate_flag(
-            key, distinct_id, groups, person_properties, group_properties
+            key, distinct_id, groups, person_properties, group_properties, device_id
         )
         flag_was_locally_evaluated = flag_value is not None
 
@@ -1594,7 +1620,13 @@ class Client(object):
                 self.flag_cache.set_cached_flag(
                     distinct_id, key, flag_result, self.flag_definition_version
                 )
-        elif not only_evaluate_locally:
+        elif only_evaluate_locally:
+            if self.feature_flags is None:
+                self.log.warning(
+                    "[FEATURE FLAGS] Local evaluation called but feature flag definitions are not loaded yet. "
+                    "Returning None. You can call load_feature_flags() to load flags explicitly."
+                )
+        else:
             try:
                 flag_details, request_id, evaluated_at, errors_while_computing = (
                     self._get_feature_flag_details_from_server(
@@ -1778,6 +1810,7 @@ class Client(object):
         groups: dict[str, str],
         person_properties: dict[str, str],
         group_properties: dict[str, str],
+        device_id: Optional[str] = None,
     ) -> Optional[FlagValue]:
         if self.feature_flags is None and self.personal_api_key:
             self.load_feature_flags()
@@ -1797,6 +1830,7 @@ class Client(object):
                         groups=groups,
                         person_properties=person_properties,
                         group_properties=group_properties,
+                        device_id=device_id,
                     )
                     self.log.debug(
                         f"Successfully computed flag locally: {key} -> {response}"
@@ -1921,10 +1955,12 @@ class Client(object):
             f"{key}_{'::null::' if response is None else str(response)}"
         )
 
-        if (
-            feature_flag_reported_key
-            not in self.distinct_ids_feature_flags_reported[distinct_id]
-        ):
+        reported_flags = self.distinct_ids_feature_flags_reported.get(distinct_id)
+        if reported_flags is None:
+            reported_flags = set()
+            self.distinct_ids_feature_flags_reported[distinct_id] = reported_flags
+
+        if feature_flag_reported_key not in reported_flags:
             properties: dict[str, Any] = {
                 "$feature_flag": key,
                 "$feature_flag_response": response,
@@ -1960,9 +1996,7 @@ class Client(object):
                 groups=groups,
                 disable_geoip=disable_geoip,
             )
-            self.distinct_ids_feature_flags_reported[distinct_id].add(
-                feature_flag_reported_key
-            )
+            reported_flags.add(feature_flag_reported_key)
 
     def get_remote_config_payload(self, key: str):
         if self.disabled:
@@ -2099,12 +2133,17 @@ class Client(object):
             )
         )
 
+        # Resolve device_id from context if not provided
+        if device_id is None:
+            device_id = get_context_device_id()
+
         response, fallback_to_flags = self._get_all_flags_and_payloads_locally(
             distinct_id,
             groups=groups,
             person_properties=person_properties,
             group_properties=group_properties,
             flag_keys_to_evaluate=flag_keys_to_evaluate,
+            device_id=device_id,
         )
 
         if fallback_to_flags and not only_evaluate_locally:
@@ -2135,6 +2174,7 @@ class Client(object):
         group_properties=None,
         warn_on_unknown_groups=False,
         flag_keys_to_evaluate: Optional[list[str]] = None,
+        device_id: Optional[str] = None,
     ) -> tuple[FlagsAndPayloads, bool]:
         person_properties = person_properties or {}
         group_properties = group_properties or {}
@@ -2164,6 +2204,7 @@ class Client(object):
                         person_properties=person_properties,
                         group_properties=group_properties,
                         warn_on_unknown_groups=warn_on_unknown_groups,
+                        device_id=device_id,
                     )
                     matched_payload = self._compute_payload_locally(
                         flag["key"], flags[flag["key"]]

posthog/consumer.py

@@ -3,8 +3,6 @@ import logging
 import time
 from threading import Thread
 
-import backoff
-
 from posthog.request import APIError, DatetimeSerializer, batch_post
 
 try:
@@ -128,29 +126,41 @@ class Consumer(Thread):
     def request(self, batch):
         """Attempt to upload the batch and retry before raising an error"""
 
-        def fatal_exception(exc):
+        def is_retryable(exc):
             if isinstance(exc, APIError):
                 # retry on server errors and client errors
-                # with 429 status code (rate limited),
+                # with 408 (request timeout) or 429 (rate limited),
                 # don't retry on other client errors
                 if exc.status == "N/A":
                     return False
-                return (400 <= exc.status < 500) and exc.status != 429
+                return not ((400 <= exc.status < 500) and exc.status not in (408, 429))
             else:
                 # retry on all other errors (eg. network)
-                return False
+                return True
 
-        @backoff.on_exception(
-            backoff.expo, Exception, max_tries=self.retries + 1, giveup=fatal_exception
-        )
-        def send_request():
-            batch_post(
-                self.api_key,
-                self.host,
-                gzip=self.gzip,
-                timeout=self.timeout,
-                batch=batch,
-                historical_migration=self.historical_migration,
-            )
+        last_exc = None
+        for attempt in range(self.retries + 1):
+            try:
+                batch_post(
+                    self.api_key,
+                    self.host,
+                    gzip=self.gzip,
+                    timeout=self.timeout,
+                    batch=batch,
+                    historical_migration=self.historical_migration,
+                )
+                return
+            except Exception as e:
+                last_exc = e
+                if not is_retryable(e):
+                    raise
+                if attempt < self.retries:
+                    # Respect Retry-After header if present, otherwise use exponential backoff
+                    retry_after = getattr(e, "retry_after", None)
+                    if retry_after and retry_after > 0:
+                        time.sleep(retry_after)
+                    else:
+                        time.sleep(min(2**attempt, 30))
 
-        send_request()
+        if last_exc:
+            raise last_exc

posthog/exception_utils.py

@@ -43,26 +43,31 @@ except ImportError:
 DEFAULT_MAX_VALUE_LENGTH = 1024
 
 DEFAULT_CODE_VARIABLES_MASK_PATTERNS = [
-    r"(?i).*password.*",
-    r"(?i).*secret.*",
-    r"(?i).*passwd.*",
-    r"(?i).*pwd.*",
-    r"(?i).*api_key.*",
-    r"(?i).*apikey.*",
-    r"(?i).*auth.*",
-    r"(?i).*credentials.*",
-    r"(?i).*privatekey.*",
-    r"(?i).*private_key.*",
-    r"(?i).*token.*",
-    r"(?i).*aws_access_key_id.*",
-    r"(?i).*_pass",
-    r"(?i)sk_.*",
-    r"(?i).*jwt.*",
+    r"(?i)password",
+    r"(?i)secret",
+    r"(?i)passwd",
+    r"(?i)pwd",
+    r"(?i)api_key",
+    r"(?i)apikey",
+    r"(?i)auth",
+    r"(?i)credentials",
+    r"(?i)privatekey",
+    r"(?i)private_key",
+    r"(?i)token",
+    r"(?i)aws_access_key_id",
+    r"(?i)_pass",
+    r"(?i)sk_",
+    r"(?i)jwt",
 ]
 
 DEFAULT_CODE_VARIABLES_IGNORE_PATTERNS = [r"^__.*"]
 
 CODE_VARIABLES_REDACTED_VALUE = "$$_posthog_redacted_based_on_masking_rules_$$"
+CODE_VARIABLES_TOO_LONG_VALUE = "$$_posthog_value_too_long_$$"
+
+_MAX_VALUE_LENGTH_FOR_PATTERN_MATCH = 5_000
+_MAX_COLLECTION_ITEMS_TO_SCAN = 100
+_REGEX_METACHARACTERS = frozenset(r"\.^$*+?{}[]|()")
 
 DEFAULT_TOTAL_VARIABLES_SIZE_LIMIT = 20 * 1024
 
@@ -928,40 +933,87 @@ def strip_string(value, max_length=None):
     )
 
 
+def _extract_plain_substring(pattern):
+    # Matches inline flag groups like (?i), (?ai), (?ims), etc. that include the 'i' flag.
+    # Python regex flags: a=ASCII, i=IGNORECASE, L=LOCALE, m=MULTILINE, s=DOTALL, u=UNICODE, x=VERBOSE
+    inline_flags = re.match(r"^\(\?[aiLmsux]*i[aiLmsux]*\)", pattern)
+    if not inline_flags:
+        return None
+    remainder = pattern[inline_flags.end() :]
+    if not remainder or any(c in _REGEX_METACHARACTERS for c in remainder):
+        return None
+    return remainder.lower()
+
+
 def _compile_patterns(patterns):
-    compiled = []
+    if not patterns:
+        return None
+    substrings = []
+    regexes = []
     for pattern in patterns:
-        try:
-            compiled.append(re.compile(pattern))
-        except Exception:
-            pass
-    return compiled
+        simple = _extract_plain_substring(pattern)
+        if simple is not None:
+            substrings.append(simple)
+        else:
+            try:
+                regexes.append(re.compile(pattern))
+            except Exception:
+                pass
+    if not substrings and not regexes:
+        return None
+    return (substrings, regexes)
 
 
 def _pattern_matches(name, patterns):
-    for pattern in patterns:
+    if patterns is None:
+        return False
+    substrings, regexes = patterns
+    if substrings:
+        name_lower = name.lower()
+        for s in substrings:
+            if s in name_lower:
+                return True
+    for pattern in regexes:
         if pattern.search(name):
             return True
     return False
 
 
-def _mask_sensitive_data(value, compiled_mask):
+def _mask_sensitive_data(value, compiled_mask, _seen=None):
     if not compiled_mask:
         return value
 
+    if isinstance(value, (dict, list, tuple)):
+        if _seen is None:
+            _seen = set()
+        obj_id = id(value)
+        if obj_id in _seen:
+            return "<circular ref>"
+        _seen.add(obj_id)
+
     if isinstance(value, dict):
+        if len(value) > _MAX_COLLECTION_ITEMS_TO_SCAN:
+            return CODE_VARIABLES_TOO_LONG_VALUE
         result = {}
         for k, v in value.items():
             key_str = str(k) if not isinstance(k, str) else k
-            if _pattern_matches(key_str, compiled_mask):
+            if len(key_str) > _MAX_VALUE_LENGTH_FOR_PATTERN_MATCH:
+                result[k] = CODE_VARIABLES_TOO_LONG_VALUE
+            elif _pattern_matches(key_str, compiled_mask):
                 result[k] = CODE_VARIABLES_REDACTED_VALUE
             else:
-                result[k] = _mask_sensitive_data(v, compiled_mask)
+                result[k] = _mask_sensitive_data(v, compiled_mask, _seen)
         return result
     elif isinstance(value, (list, tuple)):
-        masked_items = [_mask_sensitive_data(item, compiled_mask) for item in value]
+        if len(value) > _MAX_COLLECTION_ITEMS_TO_SCAN:
+            return CODE_VARIABLES_TOO_LONG_VALUE
+        masked_items = [
+            _mask_sensitive_data(item, compiled_mask, _seen) for item in value
+        ]
         return type(value)(masked_items)
     elif isinstance(value, str):
+        if len(value) > _MAX_VALUE_LENGTH_FOR_PATTERN_MATCH:
+            return CODE_VARIABLES_TOO_LONG_VALUE
         if _pattern_matches(value, compiled_mask):
             return CODE_VARIABLES_REDACTED_VALUE
         return value
@@ -982,7 +1034,9 @@ def _serialize_variable_value(value, limiter, max_length=1024, compiled_mask=Non
             limiter.add(result_size)
             return value
         elif isinstance(value, str):
-            if compiled_mask and _pattern_matches(value, compiled_mask):
+            if len(value) > _MAX_VALUE_LENGTH_FOR_PATTERN_MATCH:
+                result = CODE_VARIABLES_TOO_LONG_VALUE
+            elif compiled_mask and _pattern_matches(value, compiled_mask):
                 result = CODE_VARIABLES_REDACTED_VALUE
             else:
                 result = value

posthog/feature_flags.py

@@ -2,6 +2,7 @@ import datetime
 import hashlib
 import logging
 import re
+import warnings
 from typing import Optional
 
 from dateutil import parser
@@ -17,6 +18,30 @@ log = logging.getLogger("posthog")
 
 NONE_VALUES_ALLOWED_OPERATORS = ["is_not"]
 
+# All operators supported by match_property, grouped by category.
+EQUALITY_OPERATORS = ("exact", "is_not", "is_set", "is_not_set")
+STRING_OPERATORS = ("icontains", "not_icontains", "regex", "not_regex")
+NUMERIC_OPERATORS = ("gt", "gte", "lt", "lte")
+DATE_OPERATORS = ("is_date_before", "is_date_after")
+SEMVER_COMPARISON_OPERATORS = (
+    "semver_eq",
+    "semver_neq",
+    "semver_gt",
+    "semver_gte",
+    "semver_lt",
+    "semver_lte",
+)
+SEMVER_RANGE_OPERATORS = ("semver_tilde", "semver_caret", "semver_wildcard")
+SEMVER_OPERATORS = SEMVER_COMPARISON_OPERATORS + SEMVER_RANGE_OPERATORS
+
+PROPERTY_OPERATORS = (
+    EQUALITY_OPERATORS
+    + STRING_OPERATORS
+    + NUMERIC_OPERATORS
+    + DATE_OPERATORS
+    + SEMVER_OPERATORS
+)
+
 
 class InconclusiveMatchError(Exception):
     pass
@@ -34,18 +59,18 @@ class RequiresServerEvaluation(Exception):
     pass
 
 
-# This function takes a distinct_id and a feature flag key and returns a float between 0 and 1.
-# Given the same distinct_id and key, it'll always return the same float. These floats are
+# This function takes a bucketing value and a feature flag key and returns a float between 0 and 1.
+# Given the same bucketing value and key, it'll always return the same float. These floats are
 # uniformly distributed between 0 and 1, so if we want to show this feature to 20% of traffic
-# we can do _hash(key, distinct_id) < 0.2
-def _hash(key: str, distinct_id: str, salt: str = "") -> float:
-    hash_key = f"{key}.{distinct_id}{salt}"
+# we can do _hash(key, bucketing_value) < 0.2
+def _hash(key: str, bucketing_value: str, salt: str = "") -> float:
+    hash_key = f"{key}.{bucketing_value}{salt}"
     hash_val = int(hashlib.sha1(hash_key.encode("utf-8")).hexdigest()[:15], 16)
     return hash_val / __LONG_SCALE__
 
 
-def get_matching_variant(flag, distinct_id):
-    hash_value = _hash(flag["key"], distinct_id, salt="variant")
+def get_matching_variant(flag, bucketing_value):
+    hash_value = _hash(flag["key"], bucketing_value, salt="variant")
     for variant in variant_lookup_table(flag):
         if hash_value >= variant["value_min"] and hash_value < variant["value_max"]:
             return variant["key"]
@@ -68,7 +93,13 @@ def variant_lookup_table(feature_flag):
 
 
 def evaluate_flag_dependency(
-    property, flags_by_key, evaluation_cache, distinct_id, properties, cohort_properties
+    property,
+    flags_by_key,
+    evaluation_cache,
+    distinct_id,
+    properties,
+    cohort_properties,
+    device_id=None,
 ):
     """
     Evaluate a flag dependency property according to the dependency chain algorithm.
@@ -80,6 +111,7 @@ def evaluate_flag_dependency(
         distinct_id: The distinct ID being evaluated
         properties: Person properties for evaluation
         cohort_properties: Cohort properties for evaluation
+        device_id: The device ID for bucketing (optional)
 
     Returns:
         bool: True if all dependencies in the chain evaluate to True, False otherwise
@@ -124,13 +156,27 @@ def evaluate_flag_dependency(
                 else:
                     # Recursively evaluate the dependency
                     try:
+                        dep_flag_filters = dep_flag.get("filters") or {}
+                        dep_aggregation_group_type_index = dep_flag_filters.get(
+                            "aggregation_group_type_index"
+                        )
+                        if dep_aggregation_group_type_index is not None:
+                            # Group flags should continue bucketing by the group key
+                            # from the current evaluation context.
+                            dep_bucketing_value = distinct_id
+                        else:
+                            dep_bucketing_value = resolve_bucketing_value(
+                                dep_flag, distinct_id, device_id
+                            )
                         dep_result = match_feature_flag_properties(
                             dep_flag,
                             distinct_id,
                             properties,
-                            cohort_properties,
-                            flags_by_key,
-                            evaluation_cache,
+                            cohort_properties=cohort_properties,
+                            flags_by_key=flags_by_key,
+                            evaluation_cache=evaluation_cache,
+                            device_id=device_id,
+                            bucketing_value=dep_bucketing_value,
                         )
                         evaluation_cache[dep_flag_key] = dep_result
                     except InconclusiveMatchError as e:
@@ -215,21 +261,54 @@ def matches_dependency_value(expected_value, actual_value):
     return False
 
 
+def resolve_bucketing_value(flag, distinct_id, device_id=None):
+    """Resolve the bucketing value for a flag based on its bucketing_identifier setting.
+
+    Returns:
+        The appropriate identifier string to use for hashing/bucketing.
+
+    Raises:
+        InconclusiveMatchError: If the flag requires device_id but none was provided.
+    """
+    flag_filters = flag.get("filters") or {}
+    bucketing_identifier = flag.get("bucketing_identifier") or flag_filters.get(
+        "bucketing_identifier"
+    )
+    if bucketing_identifier == "device_id":
+        if not device_id:
+            raise InconclusiveMatchError(
+                "Flag requires device_id for bucketing but none was provided"
+            )
+        return device_id
+    return distinct_id
+
+
 def match_feature_flag_properties(
     flag,
     distinct_id,
     properties,
+    *,
     cohort_properties=None,
     flags_by_key=None,
     evaluation_cache=None,
+    device_id=None,
+    bucketing_value=None,
 ) -> FlagValue:
-    flag_conditions = (flag.get("filters") or {}).get("groups") or []
+    if bucketing_value is None:
+        warnings.warn(
+            "Calling match_feature_flag_properties() without bucketing_value is deprecated. "
+            "Pass bucketing_value explicitly. This fallback will be removed in a future major release.",
+            DeprecationWarning,
+            stacklevel=2,
+        )
+        bucketing_value = resolve_bucketing_value(flag, distinct_id, device_id)
+
+    flag_filters = flag.get("filters") or {}
+    flag_conditions = flag_filters.get("groups") or []
     is_inconclusive = False
     cohort_properties = cohort_properties or {}
     # Some filters can be explicitly set to null, which require accessing variants like so
-    flag_variants = ((flag.get("filters") or {}).get("multivariate") or {}).get(
-        "variants"
-    ) or []
+    flag_variants = (flag_filters.get("multivariate") or {}).get("variants") or []
     valid_variant_keys = [variant["key"] for variant in flag_variants]
 
     for condition in flag_conditions:
@@ -244,12 +323,14 @@ def match_feature_flag_properties(
                 cohort_properties,
                 flags_by_key,
                 evaluation_cache,
+                bucketing_value=bucketing_value,
+                device_id=device_id,
             ):
                 variant_override = condition.get("variant")
                 if variant_override and variant_override in valid_variant_keys:
                     variant = variant_override
                 else:
-                    variant = get_matching_variant(flag, distinct_id)
+                    variant = get_matching_variant(flag, bucketing_value)
                 return variant or True
         except RequiresServerEvaluation:
             # Static cohort or other missing server-side data - must fallback to API
@@ -277,6 +358,9 @@ def is_condition_match(
     cohort_properties,
     flags_by_key=None,
     evaluation_cache=None,
+    *,
+    bucketing_value,
+    device_id=None,
 ) -> bool:
     rollout_percentage = condition.get("rollout_percentage")
     if len(condition.get("properties") or []) > 0:
@@ -290,6 +374,7 @@ def is_condition_match(
                     flags_by_key,
                     evaluation_cache,
                     distinct_id,
+                    device_id=device_id,
                 )
             elif property_type == "flag":
                 matches = evaluate_flag_dependency(
@@ -299,6 +384,7 @@ def is_condition_match(
                     distinct_id,
                     properties,
                     cohort_properties,
+                    device_id=device_id,
                 )
             else:
                 matches = match_property(prop, properties)
@@ -308,9 +394,9 @@ def is_condition_match(
         if rollout_percentage is None:
             return True
 
-    if rollout_percentage is not None and _hash(feature_flag["key"], distinct_id) > (
-        rollout_percentage / 100
-    ):
+    if rollout_percentage is not None and _hash(
+        feature_flag["key"], bucketing_value
+    ) > (rollout_percentage / 100):
         return False
 
     return True
@@ -323,6 +409,9 @@ def match_property(property, property_values) -> bool:
     operator = property.get("operator") or "exact"
     value = property.get("value")
 
+    if operator not in PROPERTY_OPERATORS:
+        raise InconclusiveMatchError(f"Unknown operator {operator}")
+
     if key not in property_values:
         raise InconclusiveMatchError(
             "can't match properties without a given property value"
@@ -443,7 +532,64 @@ def match_property(property, property_values) -> bool:
                 "The date provided must be a string or date object"
             )
 
-    # if we get here, we don't know how to handle the operator
+    if operator in SEMVER_OPERATORS:
+        try:
+            override_parsed = parse_semver(override_value)
+        except (ValueError, TypeError):
+            raise InconclusiveMatchError(
+                f"Person property value '{override_value}' is not a valid semver"
+            )
+
+        if operator in SEMVER_COMPARISON_OPERATORS:
+            try:
+                flag_parsed = parse_semver(value)
+            except (ValueError, TypeError):
+                raise InconclusiveMatchError(
+                    f"Flag semver value '{value}' is not a valid semver"
+                )
+
+            if operator == "semver_eq":
+                return override_parsed == flag_parsed
+            elif operator == "semver_neq":
+                return override_parsed != flag_parsed
+            elif operator == "semver_gt":
+                return override_parsed > flag_parsed
+            elif operator == "semver_gte":
+                return override_parsed >= flag_parsed
+            elif operator == "semver_lt":
+                return override_parsed < flag_parsed
+            elif operator == "semver_lte":
+                return override_parsed <= flag_parsed
+
+        elif operator == "semver_tilde":
+            try:
+                lower, upper = _tilde_bounds(str(value))
+            except (ValueError, TypeError):
+                raise InconclusiveMatchError(
+                    f"Flag semver value '{value}' is not valid for tilde operator"
+                )
+            return lower <= override_parsed < upper
+
+        elif operator == "semver_caret":
+            try:
+                lower, upper = _caret_bounds(str(value))
+            except (ValueError, TypeError):
+                raise InconclusiveMatchError(
+                    f"Flag semver value '{value}' is not valid for caret operator"
+                )
+            return lower <= override_parsed < upper
+
+        elif operator == "semver_wildcard":
+            try:
+                lower, upper = _wildcard_bounds(str(value))
+            except (ValueError, TypeError):
+                raise InconclusiveMatchError(
+                    f"Flag semver value '{value}' is not valid for wildcard operator"
+                )
+            return lower <= override_parsed < upper
+
+    # Unreachable: all operators in PROPERTY_OPERATORS are handled above,
+    # and unknown operators are rejected at the top of this function.
     raise InconclusiveMatchError(f"Unknown operator {operator}")
 
 
@@ -454,6 +600,7 @@ def match_cohort(
     flags_by_key=None,
     evaluation_cache=None,
     distinct_id=None,
+    device_id=None,
 ) -> bool:
     # Cohort properties are in the form of property groups like this:
     # {
@@ -478,6 +625,7 @@ def match_cohort(
         flags_by_key,
         evaluation_cache,
         distinct_id,
+        device_id=device_id,
     )
 
 
@@ -488,6 +636,7 @@ def match_property_group(
     flags_by_key=None,
     evaluation_cache=None,
     distinct_id=None,
+    device_id=None,
 ) -> bool:
     if not property_group:
         return True
@@ -512,6 +661,7 @@ def match_property_group(
                     flags_by_key,
                     evaluation_cache,
                     distinct_id,
+                    device_id=device_id,
                 )
                 if property_group_type == "AND":
                     if not matches:
@@ -545,6 +695,7 @@ def match_property_group(
                         flags_by_key,
                         evaluation_cache,
                         distinct_id,
+                        device_id=device_id,
                     )
                 elif prop.get("type") == "flag":
                     matches = evaluate_flag_dependency(
@@ -554,6 +705,7 @@ def match_property_group(
                         distinct_id,
                         property_values,
                         cohort_properties,
+                        device_id=device_id,
                     )
                 else:
                     matches = match_property(prop, property_values)
@@ -618,3 +770,75 @@ def relative_date_parse_for_feature_flag_matching(
         return parsed_dt
     else:
         return None
+
+
+def parse_semver(value: str) -> tuple:
+    """Parse a semver string into a comparable (major, minor, patch) integer tuple.
+
+    Matches the behavior of the sortableSemver HogQL function:
+    - Handles v-prefix, whitespace, pre-release suffixes
+    - Defaults missing components to 0 (e.g., 1.2 -> 1.2.0)
+    Raises ValueError if parsing fails.
+    """
+    text = str(value).strip().lstrip("vV")
+    # Strip pre-release/build metadata suffix
+    text = text.split("-")[0].split("+")[0]
+    parts = text.split(".")
+
+    if not parts or not parts[0]:
+        raise ValueError("Invalid semver format")
+
+    major = int(parts[0])
+    minor = int(parts[1]) if len(parts) > 1 and parts[1] else 0
+    patch = int(parts[2]) if len(parts) > 2 and parts[2] else 0
+
+    return (major, minor, patch)
+
+
+def _tilde_bounds(value: str) -> tuple:
+    """~1.2.3 means >=1.2.3 <1.3.0 (allows patch-level changes)."""
+    major, minor, patch = parse_semver(value)
+    return (major, minor, patch), (major, minor + 1, 0)
+
+
+def _caret_bounds(value: str) -> tuple:
+    """Caret follows semver spec:
+    ^1.2.3 means >=1.2.3 <2.0.0
+    ^0.2.3 means >=0.2.3 <0.3.0
+    ^0.0.3 means >=0.0.3 <0.0.4
+    """
+    major, minor, patch = parse_semver(value)
+    lower = (major, minor, patch)
+
+    if major > 0:
+        upper = (major + 1, 0, 0)
+    elif minor > 0:
+        upper = (0, minor + 1, 0)
+    else:
+        upper = (0, 0, patch + 1)
+
+    return lower, upper
+
+
+def _wildcard_bounds(value: str) -> tuple:
+    """Wildcard matching:
+    1.* means >=1.0.0 <2.0.0
+    1.2.* means >=1.2.0 <1.3.0
+    """
+    cleaned = str(value).strip().lstrip("vV").replace("*", "").rstrip(".")
+    if not cleaned:
+        raise ValueError("Invalid wildcard pattern")
+
+    parts = [p for p in cleaned.split(".") if p]
+    if not parts:
+        raise ValueError("Invalid wildcard pattern")
+
+    if len(parts) == 1:
+        major = int(parts[0])
+        return (major, 0, 0), (major + 1, 0, 0)
+    elif len(parts) == 2:
+        major, minor = int(parts[0]), int(parts[1])
+        return (major, minor, 0), (major, minor + 1, 0)
+    else:
+        major, minor, patch = int(parts[0]), int(parts[1]), int(parts[2])
+        return (major, minor, patch), (major, minor, patch + 1)

posthog/integrations/django.py

@@ -155,7 +155,7 @@ class PosthogContextMiddleware:
         # Extract IP address
         ip_address = request.headers.get("X-Forwarded-For")
         if ip_address:
-            tags["$ip_address"] = ip_address
+            tags["$ip"] = ip_address
 
         # Extract user agent
         user_agent = request.headers.get("User-Agent")

posthog/request.py

@@ -3,7 +3,7 @@ import logging
 import re
 import socket
 from dataclasses import dataclass
-from datetime import date, datetime
+from datetime import date, datetime, timezone
 from gzip import GzipFile
 from io import BytesIO
 from typing import Any, List, Optional, Tuple, Union
@@ -235,12 +235,31 @@ def _process_response(
             )
             raise QuotaLimitError(res.status_code, "Feature flags quota limited")
         return response
+    retry_after = None
+    retry_after_header = res.headers.get("Retry-After")
+    if retry_after_header:
+        try:
+            retry_after = float(retry_after_header)
+        except (ValueError, TypeError):
+            try:
+                from email.utils import parsedate_to_datetime
+
+                retry_after = max(
+                    0.0,
+                    (
+                        parsedate_to_datetime(retry_after_header)
+                        - datetime.now(timezone.utc)
+                    ).total_seconds(),
+                )
+            except (ValueError, TypeError):
+                pass
+
     try:
         payload = res.json()
         log.debug("received response: %s", payload)
-        raise APIError(res.status_code, payload["detail"])
+        raise APIError(res.status_code, payload["detail"], retry_after=retry_after)
     except (KeyError, ValueError):
-        raise APIError(res.status_code, res.text)
+        raise APIError(res.status_code, res.text, retry_after=retry_after)
 
 
 def decide(
@@ -348,9 +367,12 @@ def get(
 
 
 class APIError(Exception):
-    def __init__(self, status: Union[int, str], message: str):
+    def __init__(
+        self, status: Union[int, str], message: str, retry_after: Optional[float] = None
+    ):
         self.message = message
         self.status = status
+        self.retry_after = retry_after
 
     def __str__(self):
         msg = "[PostHog] {0} ({1})"

posthog/test/ai/anthropic/test_anthropic.py

@@ -1,7 +1,10 @@
+import json
 from unittest.mock import patch
 
 import pytest
 
+from posthog import identify_context, new_context
+
 try:
     from anthropic.types import Message, Usage
 
@@ -305,7 +308,34 @@ def test_basic_completion(mock_client, mock_anthropic_response):
         assert props["$ai_output_tokens"] == 10
         assert props["$ai_http_status"] == 200
         assert props["foo"] == "bar"
+        assert props["$ai_tokens_source"] == "sdk"
         assert isinstance(props["$ai_latency"], float)
+        # Verify raw usage metadata is passed for backend processing
+        assert "$ai_usage" in props
+        assert props["$ai_usage"] is not None
+        # Verify it's JSON-serializable
+        json.dumps(props["$ai_usage"])
+        # Verify it has expected structure
+        assert isinstance(props["$ai_usage"], dict)
+        assert "input_tokens" in props["$ai_usage"]
+        assert "output_tokens" in props["$ai_usage"]
+
+
+def test_tokens_source_passthrough(mock_client, mock_anthropic_response):
+    with patch(
+        "anthropic.resources.Messages.create", return_value=mock_anthropic_response
+    ):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_properties={"$ai_input_tokens": 99999},
+        )
+
+        props = mock_client.capture.call_args[1]["properties"]
+        assert props["$ai_tokens_source"] == "passthrough"
+        assert props["$ai_input_tokens"] == 99999
 
 
 def test_groups(mock_client, mock_anthropic_response):
@@ -917,6 +947,17 @@ def test_streaming_with_tool_calls(mock_client, mock_anthropic_stream_with_tools
         assert props["$ai_output_tokens"] == 25
         assert props["$ai_cache_read_input_tokens"] == 5
         assert props["$ai_cache_creation_input_tokens"] == 0
+        assert props["$ai_tokens_source"] == "sdk"
+
+        # Verify raw usage is captured in streaming mode (merged from events)
+        assert "$ai_usage" in props
+        assert props["$ai_usage"] is not None
+        # Verify it's JSON-serializable
+        json.dumps(props["$ai_usage"])
+        # Verify it has expected structure (merged from message_start and message_delta)
+        assert isinstance(props["$ai_usage"], dict)
+        assert "input_tokens" in props["$ai_usage"]
+        assert "output_tokens" in props["$ai_usage"]
 
 
 def test_async_streaming_with_tool_calls(mock_client, mock_anthropic_stream_with_tools):
@@ -1263,3 +1304,99 @@ def test_async_streaming_with_web_search(
         assert props["$ai_web_search_count"] == 2
         assert props["$ai_input_tokens"] == 50
         assert props["$ai_output_tokens"] == 25
+
+
+# =======================
+# Distinct ID Context Tests
+# =======================
+
+
+def test_no_distinct_id_uses_trace_id_and_personless(
+    mock_client, mock_anthropic_response
+):
+    """When no distinct_id is provided and no outer context, trace_id is used and event is personless."""
+    with patch(
+        "anthropic.resources.Messages.create", return_value=mock_anthropic_response
+    ):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_trace_id="trace-123",
+        )
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+
+        assert call_args["distinct_id"] == "trace-123"
+        assert props["$process_person_profile"] is False
+
+
+def test_explicit_distinct_id_creates_person_profile(
+    mock_client, mock_anthropic_response
+):
+    """When posthog_distinct_id is explicitly passed, it is used and event is not personless."""
+    with patch(
+        "anthropic.resources.Messages.create", return_value=mock_anthropic_response
+    ):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="user-123",
+            posthog_trace_id="trace-123",
+        )
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+
+        assert call_args["distinct_id"] == "user-123"
+        assert (
+            "$process_person_profile" not in props
+            or props["$process_person_profile"] is not False
+        )
+
+
+def test_outer_context_distinct_id_is_used(mock_client, mock_anthropic_response):
+    """When an outer context has a distinct_id, it should be used instead of trace_id."""
+    with patch(
+        "anthropic.resources.Messages.create", return_value=mock_anthropic_response
+    ):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        with new_context():
+            identify_context("outer-user-456")
+            client.messages.create(
+                model="claude-3-opus-20240229",
+                messages=[{"role": "user", "content": "Hello"}],
+                posthog_trace_id="trace-123",
+            )
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+
+        assert call_args["distinct_id"] == "outer-user-456"
+        assert (
+            "$process_person_profile" not in props
+            or props["$process_person_profile"] is not False
+        )
+
+
+def test_explicit_distinct_id_overrides_outer_context(
+    mock_client, mock_anthropic_response
+):
+    """When both outer context and explicit posthog_distinct_id are set, explicit wins."""
+    with patch(
+        "anthropic.resources.Messages.create", return_value=mock_anthropic_response
+    ):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        with new_context():
+            identify_context("outer-user-456")
+            client.messages.create(
+                model="claude-3-opus-20240229",
+                messages=[{"role": "user", "content": "Hello"}],
+                posthog_distinct_id="explicit-user-789",
+                posthog_trace_id="trace-123",
+            )
+
+        call_args = mock_client.capture.call_args[1]
+        assert call_args["distinct_id"] == "explicit-user-789"

posthog/test/ai/gemini/test_gemini.py

@@ -1,3 +1,4 @@
+import json
 from unittest.mock import MagicMock, patch
 
 import pytest
@@ -34,6 +35,13 @@ def mock_gemini_response():
     # Ensure cache and reasoning tokens are not present (not MagicMock)
     mock_usage.cached_content_token_count = 0
     mock_usage.thoughts_token_count = 0
+    # Make model_dump() return a proper dict for serialization
+    mock_usage.model_dump.return_value = {
+        "prompt_token_count": 20,
+        "candidates_token_count": 10,
+        "cached_content_token_count": 0,
+        "thoughts_token_count": 0,
+    }
     mock_response.usage_metadata = mock_usage
 
     mock_candidate = MagicMock()
@@ -69,6 +77,13 @@ def mock_gemini_response_with_function_calls():
     mock_usage.candidates_token_count = 15
     mock_usage.cached_content_token_count = 0
     mock_usage.thoughts_token_count = 0
+    # Make model_dump() return a proper dict for serialization
+    mock_usage.model_dump.return_value = {
+        "prompt_token_count": 25,
+        "candidates_token_count": 15,
+        "cached_content_token_count": 0,
+        "thoughts_token_count": 0,
+    }
     mock_response.usage_metadata = mock_usage
 
     # Mock function call
@@ -117,6 +132,13 @@ def mock_gemini_response_function_calls_only():
     mock_usage.candidates_token_count = 12
     mock_usage.cached_content_token_count = 0
     mock_usage.thoughts_token_count = 0
+    # Make model_dump() return a proper dict for serialization
+    mock_usage.model_dump.return_value = {
+        "prompt_token_count": 30,
+        "candidates_token_count": 12,
+        "cached_content_token_count": 0,
+        "thoughts_token_count": 0,
+    }
     mock_response.usage_metadata = mock_usage
 
     # Mock function call
@@ -174,6 +196,15 @@ def test_new_client_basic_generation(
     assert props["foo"] == "bar"
     assert "$ai_trace_id" in props
     assert props["$ai_latency"] > 0
+    # Verify raw usage metadata is passed for backend processing
+    assert "$ai_usage" in props
+    assert props["$ai_usage"] is not None
+    # Verify it's JSON-serializable
+    json.dumps(props["$ai_usage"])
+    # Verify it has expected structure
+    assert isinstance(props["$ai_usage"], dict)
+    assert "prompt_token_count" in props["$ai_usage"]
+    assert "candidates_token_count" in props["$ai_usage"]
 
 
 def test_new_client_streaming_with_generate_content_stream(
@@ -810,6 +841,13 @@ def test_streaming_cache_and_reasoning_tokens(mock_client, mock_google_genai_cli
     chunk1_usage.candidates_token_count = 5
     chunk1_usage.cached_content_token_count = 30  # Cache tokens
     chunk1_usage.thoughts_token_count = 0
+    # Make model_dump() return a proper dict for serialization
+    chunk1_usage.model_dump.return_value = {
+        "prompt_token_count": 100,
+        "candidates_token_count": 5,
+        "cached_content_token_count": 30,
+        "thoughts_token_count": 0,
+    }
     chunk1.usage_metadata = chunk1_usage
 
     chunk2 = MagicMock()
@@ -819,6 +857,13 @@ def test_streaming_cache_and_reasoning_tokens(mock_client, mock_google_genai_cli
     chunk2_usage.candidates_token_count = 10
     chunk2_usage.cached_content_token_count = 30  # Same cache tokens
     chunk2_usage.thoughts_token_count = 5  # Reasoning tokens
+    # Make model_dump() return a proper dict for serialization
+    chunk2_usage.model_dump.return_value = {
+        "prompt_token_count": 100,
+        "candidates_token_count": 10,
+        "cached_content_token_count": 30,
+        "thoughts_token_count": 5,
+    }
     chunk2.usage_metadata = chunk2_usage
 
     mock_stream = iter([chunk1, chunk2])
@@ -848,6 +893,16 @@ def test_streaming_cache_and_reasoning_tokens(mock_client, mock_google_genai_cli
     assert props["$ai_cache_read_input_tokens"] == 30
     assert props["$ai_reasoning_tokens"] == 5
 
+    # Verify raw usage is captured in streaming mode (merged from chunks)
+    assert "$ai_usage" in props
+    assert props["$ai_usage"] is not None
+    # Verify it's JSON-serializable
+    json.dumps(props["$ai_usage"])
+    # Verify it has expected structure
+    assert isinstance(props["$ai_usage"], dict)
+    assert "prompt_token_count" in props["$ai_usage"]
+    assert "candidates_token_count" in props["$ai_usage"]
+
 
 def test_web_search_grounding(mock_client, mock_google_genai_client):
     """Test web search detection via grounding_metadata."""

posthog/test/ai/openai/test_openai.py

@@ -1,3 +1,4 @@
+import json
 import time
 from unittest.mock import AsyncMock, patch
 
@@ -496,6 +497,15 @@ def test_basic_completion(mock_client, mock_openai_response):
         assert props["$ai_http_status"] == 200
         assert props["foo"] == "bar"
         assert isinstance(props["$ai_latency"], float)
+        # Verify raw usage metadata is passed for backend processing
+        assert "$ai_usage" in props
+        assert props["$ai_usage"] is not None
+        # Verify it's JSON-serializable
+        json.dumps(props["$ai_usage"])
+        # Verify it has expected structure
+        assert isinstance(props["$ai_usage"], dict)
+        assert "prompt_tokens" in props["$ai_usage"]
+        assert "completion_tokens" in props["$ai_usage"]
 
 
 def test_embeddings(mock_client, mock_embedding_response):
@@ -922,6 +932,16 @@ def test_streaming_with_tool_calls(mock_client, streaming_tool_call_chunks):
         assert props["$ai_input_tokens"] == 20
         assert props["$ai_output_tokens"] == 15
 
+        # Verify raw usage is captured in streaming mode
+        assert "$ai_usage" in props
+        assert props["$ai_usage"] is not None
+        # Verify it's JSON-serializable
+        json.dumps(props["$ai_usage"])
+        # Verify it has expected structure (merged from chunks)
+        assert isinstance(props["$ai_usage"], dict)
+        assert "prompt_tokens" in props["$ai_usage"]
+        assert "completion_tokens" in props["$ai_usage"]
+
 
 # test responses api
 def test_responses_api(mock_client, mock_openai_response_with_responses_api):

posthog/test/ai/openai_agents/__init__.py

@@ -0,0 +1 @@
+# Tests for OpenAI Agents SDK integration

posthog/test/ai/openai_agents/test_processor.py

@@ -0,0 +1,810 @@
+import logging
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+try:
+    from agents.tracing.span_data import (
+        AgentSpanData,
+        CustomSpanData,
+        FunctionSpanData,
+        GenerationSpanData,
+        GuardrailSpanData,
+        HandoffSpanData,
+        ResponseSpanData,
+        SpeechSpanData,
+        TranscriptionSpanData,
+    )
+
+    from posthog.ai.openai_agents import PostHogTracingProcessor, instrument
+
+    OPENAI_AGENTS_AVAILABLE = True
+except ImportError:
+    OPENAI_AGENTS_AVAILABLE = False
+
+
+# Skip all tests if OpenAI Agents SDK is not available
+pytestmark = pytest.mark.skipif(
+    not OPENAI_AGENTS_AVAILABLE, reason="OpenAI Agents SDK is not available"
+)
+
+
+@pytest.fixture(scope="function")
+def mock_client():
+    client = MagicMock()
+    client.privacy_mode = False
+    logging.getLogger("posthog").setLevel(logging.DEBUG)
+    return client
+
+
+@pytest.fixture(scope="function")
+def processor(mock_client):
+    return PostHogTracingProcessor(
+        client=mock_client,
+        distinct_id="test-user",
+        privacy_mode=False,
+    )
+
+
+@pytest.fixture
+def mock_trace():
+    trace = MagicMock()
+    trace.trace_id = "trace_123456789"
+    trace.name = "Test Workflow"
+    trace.group_id = "group_123"
+    trace.metadata = {"key": "value"}
+    return trace
+
+
+@pytest.fixture
+def mock_span():
+    span = MagicMock()
+    span.trace_id = "trace_123456789"
+    span.span_id = "span_987654321"
+    span.parent_id = None
+    span.started_at = "2024-01-01T00:00:00Z"
+    span.ended_at = "2024-01-01T00:00:01Z"
+    span.error = None
+    return span
+
+
+class TestPostHogTracingProcessor:
+    """Tests for the PostHogTracingProcessor class."""
+
+    def test_initialization(self, mock_client):
+        """Test processor initializes correctly."""
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id="user@example.com",
+            privacy_mode=True,
+            groups={"company": "acme"},
+            properties={"env": "test"},
+        )
+
+        assert processor._client == mock_client
+        assert processor._distinct_id == "user@example.com"
+        assert processor._privacy_mode is True
+        assert processor._groups == {"company": "acme"}
+        assert processor._properties == {"env": "test"}
+
+    def test_initialization_with_callable_distinct_id(self, mock_client, mock_trace):
+        """Test processor with callable distinct_id resolver."""
+
+        def resolver(trace):
+            return trace.metadata.get("user_id", "default")
+
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id=resolver,
+        )
+
+        mock_trace.metadata = {"user_id": "resolved-user"}
+        distinct_id = processor._get_distinct_id(mock_trace)
+        assert distinct_id == "resolved-user"
+
+    def test_on_trace_start_stores_metadata(self, processor, mock_client, mock_trace):
+        """Test that on_trace_start stores metadata but does not capture an event."""
+        processor.on_trace_start(mock_trace)
+
+        mock_client.capture.assert_not_called()
+        assert mock_trace.trace_id in processor._trace_metadata
+
+    def test_on_trace_end_captures_ai_trace(self, processor, mock_client, mock_trace):
+        """Test that on_trace_end captures $ai_trace event."""
+        processor.on_trace_start(mock_trace)
+        processor.on_trace_end(mock_trace)
+
+        mock_client.capture.assert_called_once()
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_trace"
+        assert call_kwargs["distinct_id"] == "test-user"
+        assert call_kwargs["properties"]["$ai_trace_id"] == "trace_123456789"
+        assert call_kwargs["properties"]["$ai_trace_name"] == "Test Workflow"
+        assert call_kwargs["properties"]["$ai_provider"] == "openai"
+        assert call_kwargs["properties"]["$ai_framework"] == "openai-agents"
+        assert "$ai_latency" in call_kwargs["properties"]
+
+    def test_personless_mode_when_no_distinct_id(self, mock_client, mock_trace):
+        """Test that trace events use personless mode when no distinct_id is provided."""
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+        )
+
+        processor.on_trace_start(mock_trace)
+        processor.on_trace_end(mock_trace)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$process_person_profile"] is False
+        # Should fallback to trace_id as the distinct_id
+        assert call_kwargs["distinct_id"] == mock_trace.trace_id
+
+    def test_personless_mode_for_spans_when_no_distinct_id(
+        self, mock_client, mock_trace, mock_span
+    ):
+        """Test that span events use personless mode when no distinct_id is provided."""
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+        )
+
+        processor.on_trace_start(mock_trace)
+        mock_client.capture.reset_mock()
+
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$process_person_profile"] is False
+        assert call_kwargs["distinct_id"] == mock_span.trace_id
+
+    def test_personless_mode_when_callable_returns_none(
+        self, mock_client, mock_trace, mock_span
+    ):
+        """Test personless mode when callable distinct_id returns None."""
+
+        def resolver(trace):
+            return None  # Simulate no user ID available
+
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id=resolver,
+        )
+
+        processor.on_trace_start(mock_trace)
+        mock_client.capture.reset_mock()
+
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$process_person_profile"] is False
+        assert call_kwargs["distinct_id"] == mock_span.trace_id
+
+    def test_person_profile_when_distinct_id_provided(self, mock_client, mock_trace):
+        """Test that events create person profiles when distinct_id is provided."""
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id="real-user",
+        )
+
+        processor.on_trace_start(mock_trace)
+        processor.on_trace_end(mock_trace)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert "$process_person_profile" not in call_kwargs["properties"]
+
+    def test_on_trace_end_clears_metadata(self, processor, mock_client, mock_trace):
+        """Test that on_trace_end clears stored trace metadata."""
+        processor.on_trace_start(mock_trace)
+        assert mock_trace.trace_id in processor._trace_metadata
+
+        processor.on_trace_end(mock_trace)
+        assert mock_trace.trace_id not in processor._trace_metadata
+        # Also verify it captured the event
+        mock_client.capture.assert_called_once()
+
+    def test_on_span_start_tracks_time(self, processor, mock_span):
+        """Test that on_span_start records start time."""
+        processor.on_span_start(mock_span)
+        assert mock_span.span_id in processor._span_start_times
+
+    def test_generation_span_mapping(self, processor, mock_client, mock_span):
+        """Test GenerationSpanData maps to $ai_generation event."""
+        span_data = GenerationSpanData(
+            input=[{"role": "user", "content": "Hello"}],
+            output=[{"role": "assistant", "content": "Hi there!"}],
+            model="gpt-4o",
+            model_config={"temperature": 0.7, "max_tokens": 100},
+            usage={"input_tokens": 10, "output_tokens": 20},
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        mock_client.capture.assert_called_once()
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_generation"
+        assert call_kwargs["properties"]["$ai_trace_id"] == "trace_123456789"
+        assert call_kwargs["properties"]["$ai_span_id"] == "span_987654321"
+        assert call_kwargs["properties"]["$ai_provider"] == "openai"
+        assert call_kwargs["properties"]["$ai_framework"] == "openai-agents"
+        assert call_kwargs["properties"]["$ai_model"] == "gpt-4o"
+        assert call_kwargs["properties"]["$ai_input_tokens"] == 10
+        assert call_kwargs["properties"]["$ai_output_tokens"] == 20
+        assert call_kwargs["properties"]["$ai_input"] == [
+            {"role": "user", "content": "Hello"}
+        ]
+        assert call_kwargs["properties"]["$ai_output_choices"] == [
+            {"role": "assistant", "content": "Hi there!"}
+        ]
+
+    def test_generation_span_with_reasoning_tokens(
+        self, processor, mock_client, mock_span
+    ):
+        """Test GenerationSpanData includes reasoning tokens when present."""
+        span_data = GenerationSpanData(
+            model="o1-preview",
+            usage={
+                "input_tokens": 100,
+                "output_tokens": 500,
+                "reasoning_tokens": 400,
+            },
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_reasoning_tokens"] == 400
+
+    def test_function_span_mapping(self, processor, mock_client, mock_span):
+        """Test FunctionSpanData maps to $ai_span event with type=tool."""
+        span_data = FunctionSpanData(
+            name="get_weather",
+            input='{"city": "San Francisco"}',
+            output="Sunny, 72F",
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_span"
+        assert call_kwargs["properties"]["$ai_span_name"] == "get_weather"
+        assert call_kwargs["properties"]["$ai_span_type"] == "tool"
+        assert (
+            call_kwargs["properties"]["$ai_input_state"] == '{"city": "San Francisco"}'
+        )
+        assert call_kwargs["properties"]["$ai_output_state"] == "Sunny, 72F"
+
+    def test_agent_span_mapping(self, processor, mock_client, mock_span):
+        """Test AgentSpanData maps to $ai_span event with type=agent."""
+        span_data = AgentSpanData(
+            name="CustomerServiceAgent",
+            handoffs=["TechnicalAgent", "BillingAgent"],
+            tools=["search", "get_order"],
+            output_type="str",
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_span"
+        assert call_kwargs["properties"]["$ai_span_name"] == "CustomerServiceAgent"
+        assert call_kwargs["properties"]["$ai_span_type"] == "agent"
+        assert call_kwargs["properties"]["$ai_agent_handoffs"] == [
+            "TechnicalAgent",
+            "BillingAgent",
+        ]
+        assert call_kwargs["properties"]["$ai_agent_tools"] == ["search", "get_order"]
+
+    def test_handoff_span_mapping(self, processor, mock_client, mock_span):
+        """Test HandoffSpanData maps to $ai_span event with type=handoff."""
+        span_data = HandoffSpanData(
+            from_agent="TriageAgent",
+            to_agent="TechnicalAgent",
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_span"
+        assert call_kwargs["properties"]["$ai_span_type"] == "handoff"
+        assert call_kwargs["properties"]["$ai_handoff_from_agent"] == "TriageAgent"
+        assert call_kwargs["properties"]["$ai_handoff_to_agent"] == "TechnicalAgent"
+        assert (
+            call_kwargs["properties"]["$ai_span_name"]
+            == "TriageAgent -> TechnicalAgent"
+        )
+
+    def test_guardrail_span_mapping(self, processor, mock_client, mock_span):
+        """Test GuardrailSpanData maps to $ai_span event with type=guardrail."""
+        span_data = GuardrailSpanData(
+            name="ContentFilter",
+            triggered=True,
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_span"
+        assert call_kwargs["properties"]["$ai_span_name"] == "ContentFilter"
+        assert call_kwargs["properties"]["$ai_span_type"] == "guardrail"
+        assert call_kwargs["properties"]["$ai_guardrail_triggered"] is True
+
+    def test_custom_span_mapping(self, processor, mock_client, mock_span):
+        """Test CustomSpanData maps to $ai_span event with type=custom."""
+        span_data = CustomSpanData(
+            name="database_query",
+            data={"query": "SELECT * FROM users", "rows": 100},
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_span"
+        assert call_kwargs["properties"]["$ai_span_name"] == "database_query"
+        assert call_kwargs["properties"]["$ai_span_type"] == "custom"
+        assert call_kwargs["properties"]["$ai_custom_data"] == {
+            "query": "SELECT * FROM users",
+            "rows": 100,
+        }
+
+    def test_privacy_mode_redacts_content(self, mock_client, mock_span):
+        """Test that privacy_mode redacts input/output content."""
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id="test-user",
+            privacy_mode=True,
+        )
+
+        span_data = GenerationSpanData(
+            input=[{"role": "user", "content": "Secret message"}],
+            output=[{"role": "assistant", "content": "Secret response"}],
+            model="gpt-4o",
+            usage={"input_tokens": 10, "output_tokens": 20},
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        # Content should be redacted
+        assert call_kwargs["properties"]["$ai_input"] is None
+        assert call_kwargs["properties"]["$ai_output_choices"] is None
+        # Token counts should still be present
+        assert call_kwargs["properties"]["$ai_input_tokens"] == 10
+        assert call_kwargs["properties"]["$ai_output_tokens"] == 20
+
+    def test_error_handling_in_span(self, processor, mock_client, mock_span):
+        """Test that span errors are captured correctly."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+        mock_span.error = {"message": "Rate limit exceeded", "data": {"code": 429}}
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["properties"]["$ai_is_error"] is True
+        assert call_kwargs["properties"]["$ai_error"] == "Rate limit exceeded"
+
+    def test_generation_span_includes_total_tokens(
+        self, processor, mock_client, mock_span
+    ):
+        """Test that $ai_total_tokens is calculated and included."""
+        span_data = GenerationSpanData(
+            model="gpt-4o",
+            usage={"input_tokens": 100, "output_tokens": 50},
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_total_tokens"] == 150
+
+    def test_error_type_categorization_model_behavior(
+        self, processor, mock_client, mock_span
+    ):
+        """Test that ModelBehaviorError is categorized correctly."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+        mock_span.error = {
+            "message": "ModelBehaviorError: Invalid JSON output",
+            "type": "ModelBehaviorError",
+        }
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_error_type"] == "model_behavior_error"
+
+    def test_error_type_categorization_user_error(
+        self, processor, mock_client, mock_span
+    ):
+        """Test that UserError is categorized correctly."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+        mock_span.error = {"message": "UserError: Tool failed", "type": "UserError"}
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_error_type"] == "user_error"
+
+    def test_error_type_categorization_input_guardrail(
+        self, processor, mock_client, mock_span
+    ):
+        """Test that InputGuardrailTripwireTriggered is categorized correctly."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+        mock_span.error = {
+            "message": "InputGuardrailTripwireTriggered: Content blocked"
+        }
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert (
+            call_kwargs["properties"]["$ai_error_type"] == "input_guardrail_triggered"
+        )
+
+    def test_error_type_categorization_output_guardrail(
+        self, processor, mock_client, mock_span
+    ):
+        """Test that OutputGuardrailTripwireTriggered is categorized correctly."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+        mock_span.error = {
+            "message": "OutputGuardrailTripwireTriggered: Response blocked"
+        }
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert (
+            call_kwargs["properties"]["$ai_error_type"] == "output_guardrail_triggered"
+        )
+
+    def test_error_type_categorization_max_turns(
+        self, processor, mock_client, mock_span
+    ):
+        """Test that MaxTurnsExceeded is categorized correctly."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+        mock_span.error = {"message": "MaxTurnsExceeded: Agent exceeded maximum turns"}
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_error_type"] == "max_turns_exceeded"
+
+    def test_error_type_categorization_unknown(self, processor, mock_client, mock_span):
+        """Test that unknown errors are categorized as unknown."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+        mock_span.error = {"message": "Some random error occurred"}
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_error_type"] == "unknown"
+
+    def test_response_span_with_output_and_total_tokens(
+        self, processor, mock_client, mock_span
+    ):
+        """Test ResponseSpanData includes output choices and total tokens."""
+        # Create a mock response object
+        mock_response = MagicMock()
+        mock_response.id = "resp_123"
+        mock_response.model = "gpt-4o"
+        mock_response.output = [{"type": "message", "content": "Hello!"}]
+        mock_response.usage = MagicMock()
+        mock_response.usage.input_tokens = 25
+        mock_response.usage.output_tokens = 10
+
+        span_data = ResponseSpanData(
+            response=mock_response,
+            input="Hello, world!",
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_generation"
+        assert call_kwargs["properties"]["$ai_total_tokens"] == 35
+        assert call_kwargs["properties"]["$ai_output_choices"] == [
+            {"type": "message", "content": "Hello!"}
+        ]
+        assert call_kwargs["properties"]["$ai_response_id"] == "resp_123"
+
+    def test_speech_span_with_pass_through_properties(
+        self, processor, mock_client, mock_span
+    ):
+        """Test SpeechSpanData includes pass-through properties."""
+        span_data = SpeechSpanData(
+            input="Hello, how can I help you?",
+            output="base64_audio_data",
+            output_format="pcm",
+            model="tts-1",
+            model_config={"voice": "alloy", "speed": 1.0},
+            first_content_at="2024-01-01T00:00:00.500Z",
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_span"
+        assert call_kwargs["properties"]["$ai_span_type"] == "speech"
+        assert call_kwargs["properties"]["$ai_model"] == "tts-1"
+        # Pass-through properties (no $ai_ prefix)
+        assert (
+            call_kwargs["properties"]["first_content_at"] == "2024-01-01T00:00:00.500Z"
+        )
+        assert call_kwargs["properties"]["audio_output_format"] == "pcm"
+        assert call_kwargs["properties"]["model_config"] == {
+            "voice": "alloy",
+            "speed": 1.0,
+        }
+        # Text input should be captured
+        assert call_kwargs["properties"]["$ai_input"] == "Hello, how can I help you?"
+
+    def test_transcription_span_with_pass_through_properties(
+        self, processor, mock_client, mock_span
+    ):
+        """Test TranscriptionSpanData includes pass-through properties."""
+        span_data = TranscriptionSpanData(
+            input="base64_audio_data",
+            input_format="pcm",
+            output="This is the transcribed text.",
+            model="whisper-1",
+            model_config={"language": "en"},
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+
+        assert call_kwargs["event"] == "$ai_span"
+        assert call_kwargs["properties"]["$ai_span_type"] == "transcription"
+        assert call_kwargs["properties"]["$ai_model"] == "whisper-1"
+        # Pass-through properties (no $ai_ prefix)
+        assert call_kwargs["properties"]["audio_input_format"] == "pcm"
+        assert call_kwargs["properties"]["model_config"] == {"language": "en"}
+        # Transcription output should be captured
+        assert (
+            call_kwargs["properties"]["$ai_output_state"]
+            == "This is the transcribed text."
+        )
+
+    def test_latency_calculation(self, processor, mock_client, mock_span):
+        """Test that latency is calculated correctly."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+
+        with patch("time.time") as mock_time:
+            mock_time.return_value = 1000.0
+            processor.on_span_start(mock_span)
+
+            mock_time.return_value = 1001.5  # 1.5 seconds later
+            processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_latency"] == pytest.approx(1.5, rel=0.01)
+
+    def test_groups_included_in_events(self, mock_client, mock_trace, mock_span):
+        """Test that groups are included in captured events."""
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id="test-user",
+            groups={"company": "acme", "team": "engineering"},
+        )
+
+        processor.on_trace_start(mock_trace)
+        processor.on_trace_end(mock_trace)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["groups"] == {"company": "acme", "team": "engineering"}
+
+    def test_additional_properties_included(self, mock_client, mock_trace):
+        """Test that additional properties are included in events."""
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id="test-user",
+            properties={"environment": "production", "version": "1.0"},
+        )
+
+        processor.on_trace_start(mock_trace)
+        processor.on_trace_end(mock_trace)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["environment"] == "production"
+        assert call_kwargs["properties"]["version"] == "1.0"
+
+    def test_shutdown_clears_state(self, processor):
+        """Test that shutdown clears internal state."""
+        processor._span_start_times["span_1"] = 1000.0
+        processor._trace_metadata["trace_1"] = {"name": "test"}
+
+        processor.shutdown()
+
+        assert len(processor._span_start_times) == 0
+        assert len(processor._trace_metadata) == 0
+
+    def test_force_flush_calls_client_flush(self, processor, mock_client):
+        """Test that force_flush calls client.flush()."""
+        processor.force_flush()
+        mock_client.flush.assert_called_once()
+
+    def test_generation_span_with_no_usage(self, processor, mock_client, mock_span):
+        """Test GenerationSpanData with no usage data defaults to zero tokens."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_input_tokens"] == 0
+        assert call_kwargs["properties"]["$ai_output_tokens"] == 0
+        assert call_kwargs["properties"]["$ai_total_tokens"] == 0
+
+    def test_generation_span_with_partial_usage(
+        self, processor, mock_client, mock_span
+    ):
+        """Test GenerationSpanData with only input_tokens present."""
+        span_data = GenerationSpanData(
+            model="gpt-4o",
+            usage={"input_tokens": 42},
+        )
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_input_tokens"] == 42
+        assert call_kwargs["properties"]["$ai_output_tokens"] == 0
+        assert call_kwargs["properties"]["$ai_total_tokens"] == 42
+
+    def test_error_type_categorization_by_type_field_only(
+        self, processor, mock_client, mock_span
+    ):
+        """Test error categorization works when only the type field matches."""
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+        mock_span.error = {
+            "message": "Something went wrong",
+            "type": "ModelBehaviorError",
+        }
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["properties"]["$ai_error_type"] == "model_behavior_error"
+
+    def test_distinct_id_resolved_from_trace_for_spans(
+        self, mock_client, mock_trace, mock_span
+    ):
+        """Test that spans use the distinct_id resolved at trace start."""
+
+        def resolver(trace):
+            return f"user-{trace.name}"
+
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id=resolver,
+        )
+
+        # Start trace - this resolves and stores distinct_id
+        processor.on_trace_start(mock_trace)
+        mock_client.capture.reset_mock()
+
+        # End a span - should use the stored distinct_id from trace
+        span_data = GenerationSpanData(model="gpt-4o")
+        mock_span.span_data = span_data
+
+        processor.on_span_start(mock_span)
+        processor.on_span_end(mock_span)
+
+        call_kwargs = mock_client.capture.call_args[1]
+        assert call_kwargs["distinct_id"] == "user-Test Workflow"
+
+    def test_eviction_of_stale_entries(self, mock_client):
+        """Test that stale entries are evicted when max is exceeded."""
+        processor = PostHogTracingProcessor(
+            client=mock_client,
+            distinct_id="test-user",
+        )
+        processor._max_tracked_entries = 10
+
+        # Fill beyond max
+        for i in range(15):
+            processor._span_start_times[f"span_{i}"] = float(i)
+            processor._trace_metadata[f"trace_{i}"] = {"name": f"trace_{i}"}
+
+        processor._evict_stale_entries()
+
+        # Should have evicted half
+        assert len(processor._span_start_times) <= 10
+        assert len(processor._trace_metadata) <= 10
+
+
+class TestInstrumentHelper:
+    """Tests for the instrument() convenience function."""
+
+    def test_instrument_registers_processor(self, mock_client):
+        """Test that instrument() registers a processor."""
+        with patch("agents.tracing.add_trace_processor") as mock_add:
+            processor = instrument(
+                client=mock_client,
+                distinct_id="test-user",
+            )
+
+            mock_add.assert_called_once_with(processor)
+            assert isinstance(processor, PostHogTracingProcessor)
+
+    def test_instrument_with_privacy_mode(self, mock_client):
+        """Test instrument() respects privacy_mode."""
+        with patch("agents.tracing.add_trace_processor"):
+            processor = instrument(
+                client=mock_client,
+                privacy_mode=True,
+            )
+
+            assert processor._privacy_mode is True
+
+    def test_instrument_with_groups_and_properties(self, mock_client):
+        """Test instrument() accepts groups and properties."""
+        with patch("agents.tracing.add_trace_processor"):
+            processor = instrument(
+                client=mock_client,
+                groups={"company": "acme"},
+                properties={"env": "test"},
+            )
+
+            assert processor._groups == {"company": "acme"}
+            assert processor._properties == {"env": "test"}

posthog/test/ai/test_prompts.py

@@ -0,0 +1,607 @@
+import unittest
+from unittest.mock import MagicMock, patch
+
+from posthog.ai.prompts import Prompts
+
+
+class MockResponse:
+    """Mock HTTP response for testing."""
+
+    def __init__(self, json_data=None, status_code=200, ok=True):
+        self._json_data = json_data
+        self.status_code = status_code
+        self.ok = ok
+
+    def json(self):
+        if self._json_data is None:
+            raise ValueError("No JSON data")
+        return self._json_data
+
+
+class TestPrompts(unittest.TestCase):
+    """Tests for the Prompts class."""
+
+    mock_prompt_response = {
+        "id": 1,
+        "name": "test-prompt",
+        "prompt": "Hello, {{name}}! You are a helpful assistant for {{company}}.",
+        "version": 1,
+        "created_by": "user@example.com",
+        "created_at": "2024-01-01T00:00:00Z",
+        "updated_at": "2024-01-01T00:00:00Z",
+        "deleted": False,
+    }
+
+    def create_mock_posthog(
+        self,
+        personal_api_key="phx_test_key",
+        project_api_key="phc_test_key",
+        host="https://us.posthog.com",
+    ):
+        """Create a mock PostHog client."""
+        mock = MagicMock()
+        mock.personal_api_key = personal_api_key
+        mock.api_key = project_api_key
+        mock.raw_host = host
+        return mock
+
+
+class TestPromptsGet(TestPrompts):
+    """Tests for the Prompts.get() method."""
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_successfully_fetch_a_prompt(self, mock_get_session):
+        """Should successfully fetch a prompt."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.get("test-prompt")
+
+        self.assertEqual(result, self.mock_prompt_response["prompt"])
+        mock_get.assert_called_once()
+        call_args = mock_get.call_args
+        self.assertEqual(
+            call_args[0][0],
+            "https://us.posthog.com/api/environments/@current/llm_prompts/name/test-prompt/?token=phc_test_key",
+        )
+        self.assertIn("Authorization", call_args[1]["headers"])
+        self.assertEqual(
+            call_args[1]["headers"]["Authorization"], "Bearer phx_test_key"
+        )
+
+    @patch("posthog.ai.prompts._get_session")
+    @patch("posthog.ai.prompts.time.time")
+    def test_return_cached_prompt_when_fresh(self, mock_time, mock_get_session):
+        """Should return cached prompt when fresh (no API call)."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+        mock_time.return_value = 1000.0
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        # First call - fetches from API
+        result1 = prompts.get("test-prompt", cache_ttl_seconds=300)
+        self.assertEqual(result1, self.mock_prompt_response["prompt"])
+        self.assertEqual(mock_get.call_count, 1)
+
+        # Advance time by 60 seconds (still within TTL)
+        mock_time.return_value = 1060.0
+
+        # Second call - should use cache
+        result2 = prompts.get("test-prompt", cache_ttl_seconds=300)
+        self.assertEqual(result2, self.mock_prompt_response["prompt"])
+        self.assertEqual(mock_get.call_count, 1)  # No additional fetch
+
+    @patch("posthog.ai.prompts._get_session")
+    @patch("posthog.ai.prompts.time.time")
+    def test_refetch_when_cache_is_stale(self, mock_time, mock_get_session):
+        """Should refetch when cache is stale."""
+        mock_get = mock_get_session.return_value.get
+        updated_prompt_response = {
+            **self.mock_prompt_response,
+            "prompt": "Updated prompt: Hello, {{name}}!",
+        }
+
+        mock_get.side_effect = [
+            MockResponse(json_data=self.mock_prompt_response),
+            MockResponse(json_data=updated_prompt_response),
+        ]
+        mock_time.return_value = 1000.0
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        # First call - fetches from API
+        result1 = prompts.get("test-prompt", cache_ttl_seconds=60)
+        self.assertEqual(result1, self.mock_prompt_response["prompt"])
+        self.assertEqual(mock_get.call_count, 1)
+
+        # Advance time past TTL
+        mock_time.return_value = 1061.0
+
+        # Second call - should refetch
+        result2 = prompts.get("test-prompt", cache_ttl_seconds=60)
+        self.assertEqual(result2, updated_prompt_response["prompt"])
+        self.assertEqual(mock_get.call_count, 2)
+
+    @patch("posthog.ai.prompts._get_session")
+    @patch("posthog.ai.prompts.time.time")
+    @patch("posthog.ai.prompts.log")
+    def test_use_stale_cache_on_fetch_failure_with_warning(
+        self, mock_log, mock_time, mock_get_session
+    ):
+        """Should use stale cache on fetch failure with warning."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.side_effect = [
+            MockResponse(json_data=self.mock_prompt_response),
+            Exception("Network error"),
+        ]
+        mock_time.return_value = 1000.0
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        # First call - populates cache
+        result1 = prompts.get("test-prompt", cache_ttl_seconds=60)
+        self.assertEqual(result1, self.mock_prompt_response["prompt"])
+
+        # Advance time past TTL
+        mock_time.return_value = 1061.0
+
+        # Second call - should use stale cache
+        result2 = prompts.get("test-prompt", cache_ttl_seconds=60)
+        self.assertEqual(result2, self.mock_prompt_response["prompt"])
+
+        # Check warning was logged
+        mock_log.warning.assert_called()
+        warning_call = mock_log.warning.call_args
+        self.assertIn("using stale cache", warning_call[0][0])
+
+    @patch("posthog.ai.prompts._get_session")
+    @patch("posthog.ai.prompts.log")
+    def test_use_fallback_when_no_cache_and_fetch_fails_with_warning(
+        self, mock_log, mock_get_session
+    ):
+        """Should use fallback when no cache and fetch fails with warning."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.side_effect = Exception("Network error")
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        fallback = "Default system prompt."
+        result = prompts.get("test-prompt", fallback=fallback)
+
+        self.assertEqual(result, fallback)
+
+        # Check warning was logged
+        mock_log.warning.assert_called()
+        warning_call = mock_log.warning.call_args
+        self.assertIn("using fallback", warning_call[0][0])
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_throw_when_no_cache_no_fallback_and_fetch_fails(self, mock_get_session):
+        """Should throw when no cache, no fallback, and fetch fails."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.side_effect = Exception("Network error")
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        with self.assertRaises(Exception) as context:
+            prompts.get("test-prompt")
+
+        self.assertIn("Network error", str(context.exception))
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_handle_404_response(self, mock_get_session):
+        """Should handle 404 response."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(status_code=404, ok=False)
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        with self.assertRaises(Exception) as context:
+            prompts.get("nonexistent-prompt")
+
+        self.assertIn('Prompt "nonexistent-prompt" not found', str(context.exception))
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_handle_403_response(self, mock_get_session):
+        """Should handle 403 response."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(status_code=403, ok=False)
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        with self.assertRaises(Exception) as context:
+            prompts.get("restricted-prompt")
+
+        self.assertIn(
+            'Access denied for prompt "restricted-prompt"', str(context.exception)
+        )
+
+    def test_throw_when_no_personal_api_key_configured(self):
+        """Should throw when no personal_api_key is configured."""
+        posthog = self.create_mock_posthog(personal_api_key=None)
+        prompts = Prompts(posthog)
+
+        with self.assertRaises(Exception) as context:
+            prompts.get("test-prompt")
+
+        self.assertIn(
+            "personal_api_key is required to fetch prompts", str(context.exception)
+        )
+
+    def test_throw_when_no_project_api_key_configured(self):
+        """Should throw when no project_api_key is configured."""
+        posthog = self.create_mock_posthog(project_api_key=None)
+        prompts = Prompts(posthog)
+
+        with self.assertRaises(Exception) as context:
+            prompts.get("test-prompt")
+
+        self.assertIn(
+            "project_api_key is required to fetch prompts", str(context.exception)
+        )
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_throw_when_api_returns_invalid_response_format(self, mock_get_session):
+        """Should throw when API returns invalid response format."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data={"invalid": "response"})
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        with self.assertRaises(Exception) as context:
+            prompts.get("test-prompt")
+
+        self.assertIn("Invalid response format", str(context.exception))
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_use_custom_host_from_posthog_options(self, mock_get_session):
+        """Should use custom host from PostHog options."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+
+        posthog = self.create_mock_posthog(host="https://eu.posthog.com")
+        prompts = Prompts(posthog)
+
+        prompts.get("test-prompt")
+
+        call_args = mock_get.call_args
+        self.assertTrue(
+            call_args[0][0].startswith(
+                "https://eu.posthog.com/api/environments/@current/llm_prompts/name/test-prompt/?token=phc_test_key"
+            ),
+            f"Expected URL to start with 'https://eu.posthog.com/api/environments/@current/llm_prompts/name/test-prompt/?token=phc_test_key', got {call_args[0][0]}",
+        )
+
+    @patch("posthog.ai.prompts._get_session")
+    @patch("posthog.ai.prompts.time.time")
+    def test_use_default_cache_ttl_5_minutes(self, mock_time, mock_get_session):
+        """Should use default cache TTL (5 minutes) when not specified."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+        mock_time.return_value = 1000.0
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        # First call
+        prompts.get("test-prompt")
+        self.assertEqual(mock_get.call_count, 1)
+
+        # Advance time by 4 minutes (within default 5-minute TTL)
+        mock_time.return_value = 1000.0 + (4 * 60)
+
+        # Second call - should use cache
+        prompts.get("test-prompt")
+        self.assertEqual(mock_get.call_count, 1)
+
+        # Advance time past 5-minute TTL
+        mock_time.return_value = 1000.0 + (6 * 60)
+
+        # Third call - should refetch
+        prompts.get("test-prompt")
+        self.assertEqual(mock_get.call_count, 2)
+
+    @patch("posthog.ai.prompts._get_session")
+    @patch("posthog.ai.prompts.time.time")
+    def test_use_custom_default_cache_ttl_from_constructor(
+        self, mock_time, mock_get_session
+    ):
+        """Should use custom default cache TTL from constructor."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+        mock_time.return_value = 1000.0
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog, default_cache_ttl_seconds=60)
+
+        # First call
+        prompts.get("test-prompt")
+        self.assertEqual(mock_get.call_count, 1)
+
+        # Advance time past custom TTL
+        mock_time.return_value = 1061.0
+
+        # Second call - should refetch
+        prompts.get("test-prompt")
+        self.assertEqual(mock_get.call_count, 2)
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_url_encode_prompt_names_with_special_characters(self, mock_get_session):
+        """Should URL-encode prompt names with special characters."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        prompts.get("prompt with spaces/and/slashes")
+
+        call_args = mock_get.call_args
+        self.assertEqual(
+            call_args[0][0],
+            "https://us.posthog.com/api/environments/@current/llm_prompts/name/prompt%20with%20spaces%2Fand%2Fslashes/?token=phc_test_key",
+        )
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_work_with_direct_options_no_posthog_client(self, mock_get_session):
+        """Should work with direct options (no PostHog client)."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+
+        prompts = Prompts(
+            personal_api_key="phx_direct_key", project_api_key="phc_direct_key"
+        )
+
+        result = prompts.get("test-prompt")
+
+        self.assertEqual(result, self.mock_prompt_response["prompt"])
+        call_args = mock_get.call_args
+        self.assertEqual(
+            call_args[0][0],
+            "https://us.posthog.com/api/environments/@current/llm_prompts/name/test-prompt/?token=phc_direct_key",
+        )
+        self.assertEqual(
+            call_args[1]["headers"]["Authorization"], "Bearer phx_direct_key"
+        )
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_use_custom_host_from_direct_options(self, mock_get_session):
+        """Should use custom host from direct options."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+
+        prompts = Prompts(
+            personal_api_key="phx_direct_key",
+            project_api_key="phc_direct_key",
+            host="https://eu.posthog.com",
+        )
+
+        prompts.get("test-prompt")
+
+        call_args = mock_get.call_args
+        self.assertEqual(
+            call_args[0][0],
+            "https://eu.posthog.com/api/environments/@current/llm_prompts/name/test-prompt/?token=phc_direct_key",
+        )
+
+    @patch("posthog.ai.prompts._get_session")
+    @patch("posthog.ai.prompts.time.time")
+    def test_use_custom_default_cache_ttl_from_direct_options(
+        self, mock_time, mock_get_session
+    ):
+        """Should use custom default cache TTL from direct options."""
+        mock_get = mock_get_session.return_value.get
+        mock_get.return_value = MockResponse(json_data=self.mock_prompt_response)
+        mock_time.return_value = 1000.0
+
+        prompts = Prompts(
+            personal_api_key="phx_direct_key",
+            project_api_key="phc_direct_key",
+            default_cache_ttl_seconds=60,
+        )
+
+        # First call
+        prompts.get("test-prompt")
+        self.assertEqual(mock_get.call_count, 1)
+
+        # Advance time past custom TTL
+        mock_time.return_value = 1061.0
+
+        # Second call - should refetch
+        prompts.get("test-prompt")
+        self.assertEqual(mock_get.call_count, 2)
+
+
+class TestPromptsCompile(TestPrompts):
+    """Tests for the Prompts.compile() method."""
+
+    def test_replace_a_single_variable(self):
+        """Should replace a single variable."""
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.compile("Hello, {{name}}!", {"name": "World"})
+
+        self.assertEqual(result, "Hello, World!")
+
+    def test_replace_multiple_variables(self):
+        """Should replace multiple variables."""
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.compile(
+            "Hello, {{name}}! Welcome to {{company}}. Your tier is {{tier}}.",
+            {"name": "John", "company": "Acme Corp", "tier": "premium"},
+        )
+
+        self.assertEqual(
+            result, "Hello, John! Welcome to Acme Corp. Your tier is premium."
+        )
+
+    def test_handle_numbers(self):
+        """Should handle numbers."""
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.compile("You have {{count}} items.", {"count": 42})
+
+        self.assertEqual(result, "You have 42 items.")
+
+    def test_handle_booleans(self):
+        """Should handle booleans."""
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.compile("Feature enabled: {{enabled}}", {"enabled": True})
+
+        self.assertEqual(result, "Feature enabled: True")
+
+    def test_leave_unmatched_variables_unchanged(self):
+        """Should leave unmatched variables unchanged."""
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.compile(
+            "Hello, {{name}}! Your {{unknown}} is ready.", {"name": "World"}
+        )
+
+        self.assertEqual(result, "Hello, World! Your {{unknown}} is ready.")
+
+    def test_handle_prompts_with_no_variables(self):
+        """Should handle prompts with no variables."""
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.compile("You are a helpful assistant.", {})
+
+        self.assertEqual(result, "You are a helpful assistant.")
+
+    def test_handle_empty_variables_dict(self):
+        """Should handle empty variables dict."""
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.compile("Hello, {{name}}!", {})
+
+        self.assertEqual(result, "Hello, {{name}}!")
+
+    def test_handle_multiple_occurrences_of_same_variable(self):
+        """Should handle multiple occurrences of the same variable."""
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        result = prompts.compile(
+            "Hello, {{name}}! Goodbye, {{name}}!", {"name": "World"}
+        )
+
+        self.assertEqual(result, "Hello, World! Goodbye, World!")
+
+    def test_work_with_direct_options_initialization(self):
+        """Should work with direct options initialization."""
+        prompts = Prompts(
+            personal_api_key="phx_test_key", project_api_key="phc_test_key"
+        )
+
+        result = prompts.compile("Hello, {{name}}!", {"name": "World"})
+
+        self.assertEqual(result, "Hello, World!")
+
+    def test_handle_variables_with_hyphens(self):
+        """Should handle variables with hyphens."""
+        prompts = Prompts(
+            personal_api_key="phx_test_key", project_api_key="phc_test_key"
+        )
+
+        result = prompts.compile("User ID: {{user-id}}", {"user-id": "12345"})
+
+        self.assertEqual(result, "User ID: 12345")
+
+    def test_handle_variables_with_dots(self):
+        """Should handle variables with dots."""
+        prompts = Prompts(
+            personal_api_key="phx_test_key", project_api_key="phc_test_key"
+        )
+
+        result = prompts.compile("Company: {{company.name}}", {"company.name": "Acme"})
+
+        self.assertEqual(result, "Company: Acme")
+
+
+class TestPromptsClearCache(TestPrompts):
+    """Tests for the Prompts.clear_cache() method."""
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_clear_a_specific_prompt_from_cache(self, mock_get_session):
+        """Should clear a specific prompt from cache."""
+        mock_get = mock_get_session.return_value.get
+        other_prompt_response = {**self.mock_prompt_response, "name": "other-prompt"}
+
+        mock_get.side_effect = [
+            MockResponse(json_data=self.mock_prompt_response),
+            MockResponse(json_data=other_prompt_response),
+            MockResponse(json_data=self.mock_prompt_response),
+        ]
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        # Populate cache with two prompts
+        prompts.get("test-prompt")
+        prompts.get("other-prompt")
+        self.assertEqual(mock_get.call_count, 2)
+
+        # Clear only test-prompt
+        prompts.clear_cache("test-prompt")
+
+        # test-prompt should be refetched
+        prompts.get("test-prompt")
+        self.assertEqual(mock_get.call_count, 3)
+
+        # other-prompt should still be cached
+        prompts.get("other-prompt")
+        self.assertEqual(mock_get.call_count, 3)
+
+    @patch("posthog.ai.prompts._get_session")
+    def test_clear_all_prompts_from_cache(self, mock_get_session):
+        """Should clear all prompts from cache when no name is provided."""
+        mock_get = mock_get_session.return_value.get
+        other_prompt_response = {**self.mock_prompt_response, "name": "other-prompt"}
+
+        mock_get.side_effect = [
+            MockResponse(json_data=self.mock_prompt_response),
+            MockResponse(json_data=other_prompt_response),
+            MockResponse(json_data=self.mock_prompt_response),
+            MockResponse(json_data=other_prompt_response),
+        ]
+
+        posthog = self.create_mock_posthog()
+        prompts = Prompts(posthog)
+
+        # Populate cache with two prompts
+        prompts.get("test-prompt")
+        prompts.get("other-prompt")
+        self.assertEqual(mock_get.call_count, 2)
+
+        # Clear all cache
+        prompts.clear_cache()
+
+        # Both prompts should be refetched
+        prompts.get("test-prompt")
+        prompts.get("other-prompt")
+        self.assertEqual(mock_get.call_count, 4)
+
+
+if __name__ == "__main__":
+    unittest.main()

posthog/test/ai/test_sanitization.py

@@ -69,6 +69,25 @@ class TestSanitization(unittest.TestCase):
         )
         self.assertEqual(result[0]["content"][1]["image_url"]["detail"], "high")
 
+    def test_sanitize_openai_input_image(self):
+        input_data = [
+            {
+                "role": "user",
+                "content": [
+                    {
+                        "type": "input_image",
+                        "image_url": self.sample_base64_image,
+                    }
+                ],
+            }
+        ]
+
+        result = sanitize_openai(input_data)
+
+        self.assertEqual(
+            result[0]["content"][0]["image_url"], REDACTED_IMAGE_PLACEHOLDER
+        )
+
     def test_sanitize_openai_preserves_regular_urls(self):
         input_data = [
             {

posthog/test/ai/test_tokens_source.py

@@ -0,0 +1,62 @@
+from parameterized import parameterized
+
+from posthog.ai.utils import _get_tokens_source
+
+
+@parameterized.expand(
+    [
+        ("no_posthog_properties", {"$ai_input_tokens": 100}, None, "sdk"),
+        ("empty_posthog_properties", {"$ai_input_tokens": 100}, {}, "sdk"),
+        (
+            "unrelated_posthog_properties",
+            {"$ai_input_tokens": 100},
+            {"foo": "bar"},
+            "sdk",
+        ),
+        (
+            "override_input_tokens",
+            {"$ai_input_tokens": 100},
+            {"$ai_input_tokens": 999},
+            "passthrough",
+        ),
+        (
+            "override_output_tokens",
+            {"$ai_output_tokens": 50},
+            {"$ai_output_tokens": 999},
+            "passthrough",
+        ),
+        (
+            "override_total_tokens",
+            {"$ai_input_tokens": 100},
+            {"$ai_total_tokens": 999},
+            "passthrough",
+        ),
+        (
+            "override_cache_read",
+            {"$ai_input_tokens": 100},
+            {"$ai_cache_read_input_tokens": 500},
+            "passthrough",
+        ),
+        (
+            "override_cache_creation",
+            {"$ai_input_tokens": 100},
+            {"$ai_cache_creation_input_tokens": 200},
+            "passthrough",
+        ),
+        (
+            "override_reasoning_tokens",
+            {"$ai_input_tokens": 100},
+            {"$ai_reasoning_tokens": 300},
+            "passthrough",
+        ),
+        (
+            "mixed_override_and_custom",
+            {"$ai_input_tokens": 100},
+            {"$ai_input_tokens": 999, "custom_key": "value"},
+            "passthrough",
+        ),
+    ]
+)
+def test_get_tokens_source(name, sdk_tags, posthog_properties, expected):
+    result = _get_tokens_source(sdk_tags, posthog_properties)
+    assert result == expected

posthog/test/test_client.py

@@ -479,6 +479,20 @@ class TestClient(unittest.TestCase):
             self.assertEqual(client.cohorts, {})
             self.assertIn("PostHog feature flags quota limited", logs.output[0])
 
+    @mock.patch("posthog.client.get")
+    def test_load_feature_flags_unauthorized(self, patch_get):
+        patch_get.side_effect = APIError(401, "Unauthorized")
+
+        client = Client(FAKE_TEST_API_KEY, personal_api_key="test")
+        with self.assertLogs("posthog", level="ERROR") as logs:
+            client._load_feature_flags()
+
+            self.assertEqual(client.feature_flags, [])
+            self.assertEqual(client.feature_flags_by_key, {})
+            self.assertEqual(client.group_type_mapping, {})
+            self.assertEqual(client.cohorts, {})
+            self.assertIn("please set a valid personal_api_key", logs.output[0])
+
     @mock.patch("posthog.client.flags")
     def test_dont_override_capture_with_local_flags(self, patch_flags):
         patch_flags.return_value = {"featureFlags": {"beta-feature": "random-variant"}}

posthog/test/test_consumer.py

@@ -167,6 +167,63 @@ class TestConsumer(unittest.TestCase):
             q.join()
             self.assertEqual(mock_post.call_count, 2)
 
+    def test_request_sleeps_with_retry_after(self) -> None:
+        error = APIError(429, "Too Many Requests", retry_after=5.0)
+        call_count = [0]
+
+        def mock_post(*args: Any, **kwargs: Any) -> None:
+            call_count[0] += 1
+            if call_count[0] <= 1:
+                raise error
+
+        consumer = Consumer(None, TEST_API_KEY, retries=3)
+        with (
+            mock.patch("posthog.consumer.batch_post", side_effect=mock_post),
+            mock.patch("posthog.consumer.time.sleep") as mock_sleep,
+        ):
+            consumer.request([_track_event()])
+            mock_sleep.assert_called_once_with(5.0)
+
+    def test_request_uses_exponential_backoff_without_retry_after(self) -> None:
+        error = APIError(503, "Service Unavailable")
+        call_count = [0]
+
+        def mock_post(*args: Any, **kwargs: Any) -> None:
+            call_count[0] += 1
+            if call_count[0] <= 3:
+                raise error
+
+        consumer = Consumer(None, TEST_API_KEY, retries=3)
+        with (
+            mock.patch("posthog.consumer.batch_post", side_effect=mock_post),
+            mock.patch("posthog.consumer.time.sleep") as mock_sleep,
+        ):
+            consumer.request([_track_event()])
+            self.assertEqual(
+                mock_sleep.call_args_list,
+                [
+                    mock.call(1),  # 2^0
+                    mock.call(2),  # 2^1
+                    mock.call(4),  # 2^2
+                ],
+            )
+
+    def test_request_retries_on_408(self) -> None:
+        call_count = [0]
+
+        def mock_post(*args: Any, **kwargs: Any) -> None:
+            call_count[0] += 1
+            if call_count[0] <= 1:
+                raise APIError(408, "Request Timeout")
+
+        consumer = Consumer(None, TEST_API_KEY, retries=3)
+        with (
+            mock.patch("posthog.consumer.batch_post", side_effect=mock_post),
+            mock.patch("posthog.consumer.time.sleep"),
+        ):
+            consumer.request([_track_event()])
+            self.assertEqual(call_count[0], 2)
+
     @parameterized.expand(
         [
             ("on_error_succeeds", False),

posthog/test/test_exception_capture.py

@@ -450,3 +450,240 @@ def test_code_variables_repr_fallback(tmpdir):
     assert "<CustomReprClass: custom representation>" in output
     assert "<lambda>" in output
     assert "<function trigger_error at" in output
+
+
+def test_code_variables_too_long_string_value_replaced(tmpdir):
+    app = tmpdir.join("app.py")
+    app.write(
+        dedent(
+            """
+    import os
+    from posthog import Posthog
+
+    posthog = Posthog(
+        'phc_x',
+        host='https://eu.i.posthog.com',
+        debug=True,
+        enable_exception_autocapture=True,
+        capture_exception_code_variables=True,
+        project_root=os.path.dirname(os.path.abspath(__file__))
+    )
+
+    def trigger_error():
+        short_value = "I am short"
+        long_value = "x" * 20000
+        long_blob = "password_" + "a" * 20000
+
+        1/0
+
+    trigger_error()
+    """
+        )
+    )
+
+    with pytest.raises(subprocess.CalledProcessError) as excinfo:
+        subprocess.check_output([sys.executable, str(app)], stderr=subprocess.STDOUT)
+
+    output = excinfo.value.output.decode("utf-8")
+
+    assert "ZeroDivisionError" in output
+    assert "code_variables" in output
+
+    assert "'short_value': 'I am short'" in output
+
+    assert "$$_posthog_value_too_long_$$" in output
+
+    assert "'long_blob': '$$_posthog_value_too_long_$$'" in output
+
+
+def test_code_variables_too_long_string_in_nested_dict(tmpdir):
+    app = tmpdir.join("app.py")
+    app.write(
+        dedent(
+            """
+    import os
+    from posthog import Posthog
+
+    posthog = Posthog(
+        'phc_x',
+        host='https://eu.i.posthog.com',
+        debug=True,
+        enable_exception_autocapture=True,
+        capture_exception_code_variables=True,
+        project_root=os.path.dirname(os.path.abspath(__file__))
+    )
+
+    def trigger_error():
+        my_data = {
+            "short_key": "short_val",
+            "long_key": "y" * 20000,
+            "nested": {
+                "deep_long": "z" * 20000,
+                "deep_short": "ok",
+            },
+        }
+
+        1/0
+
+    trigger_error()
+    """
+        )
+    )
+
+    with pytest.raises(subprocess.CalledProcessError) as excinfo:
+        subprocess.check_output([sys.executable, str(app)], stderr=subprocess.STDOUT)
+
+    output = excinfo.value.output.decode("utf-8")
+
+    assert "ZeroDivisionError" in output
+    assert "code_variables" in output
+
+    assert "short_val" in output
+    assert "ok" in output
+
+    assert "$$_posthog_value_too_long_$$" in output
+    assert "y" * 1000 not in output
+    assert "z" * 1000 not in output
+
+
+def test_mask_sensitive_data_too_long_dict_key():
+    from posthog.exception_utils import (
+        CODE_VARIABLES_TOO_LONG_VALUE,
+        _compile_patterns,
+        _mask_sensitive_data,
+    )
+
+    compiled_mask = _compile_patterns([r"(?i)password"])
+
+    result = _mask_sensitive_data(
+        {
+            "short": "visible",
+            "k" * 20000: "hidden_val",
+            "password": "secret",
+        },
+        compiled_mask,
+    )
+
+    assert result["short"] == "visible"
+    # This then gets shortened by the JSON truncation at 1024 chars anyways so no worries
+    assert result["k" * 20000] == CODE_VARIABLES_TOO_LONG_VALUE
+    assert result["password"] == "$$_posthog_redacted_based_on_masking_rules_$$"
+
+
+def test_mask_sensitive_data_circular_ref():
+    from posthog.exception_utils import _compile_patterns, _mask_sensitive_data
+
+    compiled_mask = _compile_patterns([r"(?i)password"])
+
+    # Circular dict
+    circular_dict = {"key": "value"}
+    circular_dict["self"] = circular_dict
+
+    result = _mask_sensitive_data(circular_dict, compiled_mask)
+    assert result["key"] == "value"
+    assert result["self"] == "<circular ref>"
+
+    # Circular list
+    circular_list = ["item"]
+    circular_list.append(circular_list)
+
+    result = _mask_sensitive_data(circular_list, compiled_mask)
+    assert result[0] == "item"
+    assert result[1] == "<circular ref>"
+
+
+def test_compile_patterns_fast_path_and_regex_fallback():
+    from posthog.exception_utils import _compile_patterns, _pattern_matches
+
+    # Simple case-insensitive patterns should become substrings
+    simple_only = _compile_patterns([r"(?i)password", r"(?i)token", r"(?i)jwt"])
+    substrings, regexes = simple_only
+    assert substrings == ["password", "token", "jwt"]
+    assert regexes == []
+
+    assert _pattern_matches("my_password_var", simple_only) is True
+    assert _pattern_matches("MY_TOKEN", simple_only) is True
+    assert _pattern_matches("safe_variable", simple_only) is False
+
+    # Complex regex patterns should stay as compiled regexes
+    complex_only = _compile_patterns([r"^__.*", r"\d{3,}", r"^sk_live_"])
+    substrings, regexes = complex_only
+    assert substrings == []
+    assert len(regexes) == 3
+
+    assert _pattern_matches("__dunder", complex_only) is True
+    assert _pattern_matches("has_999_numbers", complex_only) is True
+    assert _pattern_matches("sk_live_abc123", complex_only) is True
+    assert _pattern_matches("normal_var", complex_only) is False
+
+    # Mixed: simple substrings + complex regexes together
+    mixed = _compile_patterns(
+        [
+            r"(?i)secret",  # simple
+            r"(?i)api_key",  # simple
+            r"^__.*",  # regex
+            r"\btoken_\w+",  # regex
+        ]
+    )
+    substrings, regexes = mixed
+    assert substrings == ["secret", "api_key"]
+    assert len(regexes) == 2
+
+    # Substring matches
+    assert _pattern_matches("my_secret", mixed) is True
+    assert _pattern_matches("API_KEY_VALUE", mixed) is True
+
+    # Regex matches
+    assert _pattern_matches("__private", mixed) is True
+    assert _pattern_matches("token_abc", mixed) is True
+
+    # No match
+    assert _pattern_matches("safe_var", mixed) is False
+
+
+def test_mask_sensitive_data_large_dict_replaced():
+    from posthog.exception_utils import (
+        CODE_VARIABLES_TOO_LONG_VALUE,
+        _compile_patterns,
+        _mask_sensitive_data,
+    )
+
+    compiled_mask = _compile_patterns([r"(?i)password"])
+
+    large_dict = {f"key_{i}": f"value_{i}" for i in range(300)}
+
+    result = _mask_sensitive_data(large_dict, compiled_mask)
+
+    assert result == CODE_VARIABLES_TOO_LONG_VALUE
+
+
+def test_mask_sensitive_data_large_list_replaced():
+    from posthog.exception_utils import (
+        CODE_VARIABLES_TOO_LONG_VALUE,
+        _compile_patterns,
+        _mask_sensitive_data,
+    )
+
+    compiled_mask = _compile_patterns([r"(?i)password"])
+
+    large_list = [f"item_{i}" for i in range(300)]
+
+    result = _mask_sensitive_data(large_list, compiled_mask)
+
+    assert result == CODE_VARIABLES_TOO_LONG_VALUE
+
+
+def test_mask_sensitive_data_large_tuple_replaced():
+    from posthog.exception_utils import (
+        CODE_VARIABLES_TOO_LONG_VALUE,
+        _compile_patterns,
+        _mask_sensitive_data,
+    )
+
+    compiled_mask = _compile_patterns([r"(?i)password"])
+
+    large_tuple = tuple(f"item_{i}" for i in range(300))
+
+    result = _mask_sensitive_data(large_tuple, compiled_mask)
+
+    assert result == CODE_VARIABLES_TOO_LONG_VALUE

posthog/test/test_feature_flags.py

@@ -233,6 +233,27 @@ class TestLocalEvaluation(unittest.TestCase):
 
         self.assertEqual(patch_flags.call_count, 1)
 
+    def test_group_flag_is_inconclusive_when_group_properties_missing(self):
+        feature_flag = {
+            "id": 1,
+            "name": "Group Flag Without Property Filters",
+            "key": "group-flag-no-props",
+            "active": True,
+            "filters": {
+                "aggregation_group_type_index": 0,
+                "groups": [{"properties": [], "rollout_percentage": 100}],
+            },
+        }
+        self.client.group_type_mapping = {"0": "company"}
+
+        with self.assertRaises(InconclusiveMatchError):
+            self.client._compute_flag_locally(
+                feature_flag,
+                "some-distinct-id",
+                groups={"company": "acme"},
+                group_properties={},
+            )
+
     @mock.patch("posthog.client.flags")
     @mock.patch("posthog.client.get")
     def test_flag_with_complex_definition(self, patch_get, patch_flags):
@@ -2513,7 +2534,10 @@ class TestLocalEvaluation(unittest.TestCase):
         self.assertIsNone(client._flags_etag)
         self.assertEqual(client.feature_flags[0]["key"], "flag-v2")
 
-    def test_load_feature_flags_wrong_key(self):
+    @mock.patch("posthog.client.Poller")
+    @mock.patch("posthog.client.get")
+    def test_load_feature_flags_wrong_key(self, patch_get, _patch_poll):
+        patch_get.side_effect = APIError(401, "Unauthorized")
         client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
 
         with self.assertLogs("posthog", level="ERROR") as logs:
@@ -3220,6 +3244,541 @@ class TestLocalEvaluation(unittest.TestCase):
         # Verify API was called (fallback occurred)
         self.assertEqual(patch_flags.call_count, 1)
 
+    @mock.patch("posthog.client.flags")
+    def test_device_id_bucketing_uses_device_id_for_hash(self, patch_flags):
+        """
+        When a flag has bucketing_identifier: "device_id", the device_id should be
+        used for hashing instead of distinct_id.
+        """
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        # This flag uses device_id for bucketing
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "device-bucketed-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            }
+        ]
+
+        # Same distinct_id with different device_ids should produce different results
+        # (based on rollout percentage, we check consistency)
+        result1 = client.get_feature_flag(
+            "device-bucketed-flag", "user-123", device_id="device-A"
+        )
+        result2 = client.get_feature_flag(
+            "device-bucketed-flag", "user-123", device_id="device-A"
+        )
+
+        # Same device_id should give consistent results
+        self.assertEqual(result1, result2)
+
+        # No API fallback should occur
+        self.assertEqual(patch_flags.call_count, 0)
+
+    def test_match_feature_flag_properties_without_bucketing_value_is_deprecated(
+        self,
+    ):
+        """
+        match_feature_flag_properties should preserve backward compatibility when
+        bucketing_value is omitted, while warning about deprecation.
+        """
+        from posthog.feature_flags import match_feature_flag_properties
+
+        flag = {
+            "id": 1,
+            "key": "device-bucketed-flag",
+            "active": True,
+            "filters": {
+                "bucketing_identifier": "device_id",
+                "groups": [
+                    {
+                        "properties": [],
+                        "rollout_percentage": 100,
+                    }
+                ],
+            },
+        }
+
+        with self.assertWarnsRegex(
+            DeprecationWarning, "without bucketing_value is deprecated"
+        ):
+            result = match_feature_flag_properties(
+                flag,
+                "user-123",
+                {},
+                device_id="device-123",
+            )
+
+        self.assertTrue(result)
+
+    @mock.patch("posthog.client.flags")
+    def test_device_id_bucketing_same_device_different_users_same_result(
+        self, patch_flags
+    ):
+        """
+        When a flag uses device_id bucketing, different distinct_ids with the same
+        device_id should get the same result.
+        """
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "device-bucketed-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 50,
+                        }
+                    ],
+                },
+            }
+        ]
+
+        # Different distinct_ids with the same device_id should get the same result
+        result1 = client.get_feature_flag(
+            "device-bucketed-flag", "user-A", device_id="shared-device"
+        )
+        result2 = client.get_feature_flag(
+            "device-bucketed-flag", "user-B", device_id="shared-device"
+        )
+        result3 = client.get_feature_flag(
+            "device-bucketed-flag", "user-C", device_id="shared-device"
+        )
+
+        # All should be the same since device_id is the same
+        self.assertEqual(result1, result2)
+        self.assertEqual(result2, result3)
+
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_device_id_bucketing_fallback_when_device_id_missing(self, patch_flags):
+        """
+        When a flag requires device_id for bucketing but none is provided,
+        it should fallback to server evaluation.
+        """
+        patch_flags.return_value = {"featureFlags": {"device-bucketed-flag": True}}
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "device-bucketed-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            }
+        ]
+
+        # No device_id provided - should fallback to API
+        result = client.get_feature_flag("device-bucketed-flag", "user-123")
+
+        self.assertTrue(result)
+        # API should have been called
+        self.assertEqual(patch_flags.call_count, 1)
+
+    @mock.patch("posthog.client.flags")
+    def test_device_id_bucketing_returns_none_when_only_evaluate_locally_and_no_device_id(
+        self, patch_flags
+    ):
+        """
+        When only_evaluate_locally=True and device_id is required but missing,
+        should return None instead of falling back to API.
+        """
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "device-bucketed-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            }
+        ]
+
+        # No device_id + only_evaluate_locally should return None
+        result = client.get_feature_flag(
+            "device-bucketed-flag", "user-123", only_evaluate_locally=True
+        )
+
+        self.assertIsNone(result)
+        # API should NOT have been called
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_default_bucketing_identifier_uses_distinct_id(self, patch_flags):
+        """
+        When bucketing_identifier is not set or is 'distinct_id', should use
+        distinct_id for hashing (default behavior).
+        """
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        # Flag without bucketing_identifier (defaults to distinct_id)
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "normal-flag",
+                "active": True,
+                "filters": {
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 50,
+                        }
+                    ],
+                },
+            }
+        ]
+
+        # Different distinct_ids should potentially produce different results
+        # but same distinct_id should produce same result
+        result1 = client.get_feature_flag("normal-flag", "user-A")
+        result2 = client.get_feature_flag("normal-flag", "user-A")
+
+        self.assertEqual(result1, result2)
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_device_id_bucketing_with_multivariate_flag(self, patch_flags):
+        """
+        Multivariate flag variant selection should use device_id when
+        bucketing_identifier is set to device_id.
+        """
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "multivariate-device-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                    "multivariate": {
+                        "variants": [
+                            {"key": "control", "rollout_percentage": 50},
+                            {"key": "test", "rollout_percentage": 50},
+                        ]
+                    },
+                },
+            }
+        ]
+
+        # Same device_id should give same variant
+        result1 = client.get_feature_flag(
+            "multivariate-device-flag", "user-A", device_id="device-1"
+        )
+        result2 = client.get_feature_flag(
+            "multivariate-device-flag", "user-B", device_id="device-1"
+        )
+
+        # Both should get the same variant because device_id is the same
+        self.assertEqual(result1, result2)
+        self.assertIn(result1, ["control", "test"])
+
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_device_id_bucketing_from_context(self, patch_flags):
+        """
+        When device_id is not passed as a parameter but is set in the context,
+        it should be resolved from context.
+        """
+        from posthog.contexts import new_context, set_context_device_id
+
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "device-bucketed-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            }
+        ]
+
+        # Set device_id in context
+        with new_context():
+            set_context_device_id("context-device-id")
+            result = client.get_feature_flag("device-bucketed-flag", "user-123")
+
+        # Should evaluate locally using the context device_id
+        self.assertTrue(result)
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_group_flags_ignore_bucketing_identifier(self, patch_flags):
+        """
+        Group flags should continue to use the group identifier for hashing,
+        regardless of the bucketing_identifier setting.
+        """
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "group-flag",
+                "active": True,
+                "filters": {
+                    "aggregation_group_type_index": 0,
+                    "bucketing_identifier": "device_id",  # Should be ignored for group flags
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            }
+        ]
+        client.group_type_mapping = {"0": "company"}
+
+        # Even with bucketing_identifier set to device_id, group flag should use group identifier
+        result = client.get_feature_flag(
+            "group-flag",
+            "user-123",
+            groups={"company": "acme-inc"},
+            device_id="some-device",
+        )
+
+        self.assertTrue(result)
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_group_flag_dependency_receives_device_id(self, patch_flags):
+        """
+        Group flag dependency evaluation should receive device_id so dependent
+        device_id-bucketed flags can be evaluated locally.
+        """
+        patch_flags.return_value = {"featureFlags": {"group-parent-flag": "from-api"}}
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "device-dependent-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            },
+            {
+                "id": 2,
+                "key": "group-parent-flag",
+                "active": True,
+                "filters": {
+                    "aggregation_group_type_index": 0,
+                    "groups": [
+                        {
+                            "properties": [
+                                {
+                                    "key": "device-dependent-flag",
+                                    "operator": "flag_evaluates_to",
+                                    "value": True,
+                                    "type": "flag",
+                                    "dependency_chain": ["device-dependent-flag"],
+                                }
+                            ],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            },
+        ]
+        client.group_type_mapping = {"0": "company"}
+
+        result = client.get_feature_flag(
+            "group-parent-flag",
+            "user-123",
+            groups={"company": "acme-inc"},
+            device_id="device-123",
+        )
+
+        self.assertTrue(result)
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_group_flag_dependency_ignores_device_id_bucketing_identifier(
+        self, patch_flags
+    ):
+        """
+        Group flag dependencies should keep bucketing by group key, even when
+        the dependent group flag has bucketing_identifier set to device_id.
+        """
+        patch_flags.return_value = {"featureFlags": {"parent-group-flag": "from-api"}}
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "child-group-flag",
+                "active": True,
+                "filters": {
+                    "aggregation_group_type_index": 0,
+                    "bucketing_identifier": "device_id",
+                    "groups": [
+                        {
+                            "properties": [],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            },
+            {
+                "id": 2,
+                "key": "parent-group-flag",
+                "active": True,
+                "filters": {
+                    "aggregation_group_type_index": 0,
+                    "groups": [
+                        {
+                            "properties": [
+                                {
+                                    "key": "child-group-flag",
+                                    "operator": "flag_evaluates_to",
+                                    "value": True,
+                                    "type": "flag",
+                                    "dependency_chain": ["child-group-flag"],
+                                }
+                            ],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            },
+        ]
+        client.group_type_mapping = {"0": "company"}
+
+        result = client.get_feature_flag(
+            "parent-group-flag",
+            "user-123",
+            groups={"company": "acme-inc"},
+        )
+
+        self.assertTrue(result)
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_get_all_flags_with_device_id_bucketing(self, patch_flags):
+        """
+        get_all_flags_and_payloads should properly handle flags with device_id bucketing.
+        """
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "normal-flag",
+                "active": True,
+                "filters": {
+                    "groups": [{"properties": [], "rollout_percentage": 100}],
+                },
+            },
+            {
+                "id": 2,
+                "key": "device-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [{"properties": [], "rollout_percentage": 100}],
+                },
+            },
+        ]
+
+        # With device_id provided, both flags should be evaluated locally
+        result = client.get_all_flags("user-123", device_id="my-device")
+
+        self.assertEqual(result["normal-flag"], True)
+        self.assertEqual(result["device-flag"], True)
+        self.assertEqual(patch_flags.call_count, 0)
+
+    @mock.patch("posthog.client.flags")
+    def test_get_all_flags_fallback_when_device_id_missing_for_some_flags(
+        self, patch_flags
+    ):
+        """
+        When some flags require device_id but it's not provided, those flags
+        should trigger fallback while others can be evaluated locally.
+        """
+        patch_flags.return_value = {
+            "featureFlags": {"normal-flag": True, "device-flag": "from-api"}
+        }
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "key": "normal-flag",
+                "active": True,
+                "filters": {
+                    "groups": [{"properties": [], "rollout_percentage": 100}],
+                },
+            },
+            {
+                "id": 2,
+                "key": "device-flag",
+                "active": True,
+                "filters": {
+                    "bucketing_identifier": "device_id",
+                    "groups": [{"properties": [], "rollout_percentage": 100}],
+                },
+            },
+        ]
+
+        # Without device_id, device-flag can't be evaluated locally
+        client.get_all_flags("user-123")
+
+        # Should fallback to API for all flags when any can't be evaluated locally
+        self.assertEqual(patch_flags.call_count, 1)
+
 
 class TestMatchProperties(unittest.TestCase):
     def property(self, key, value, operator=None):
@@ -3639,6 +4198,207 @@ class TestMatchProperties(unittest.TestCase):
         with self.assertRaises(InconclusiveMatchError):
             self.assertFalse(match_property(property_k, {"key": "random"}))
 
+    def test_match_properties_semver_eq(self):
+        prop = self.property(key="version", value="1.2.3", operator="semver_eq")
+        self.assertTrue(match_property(prop, {"version": "1.2.3"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.4"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.2"}))
+        self.assertFalse(match_property(prop, {"version": "2.0.0"}))
+
+        # Pre-release suffix is stripped for comparison
+        self.assertTrue(match_property(prop, {"version": "1.2.3-alpha.1"}))
+
+        # Partial versions default missing parts to 0
+        prop_partial = self.property(key="version", value="1.2", operator="semver_eq")
+        self.assertTrue(match_property(prop_partial, {"version": "1.2.0"}))
+        self.assertFalse(match_property(prop_partial, {"version": "1.2.1"}))
+
+    def test_match_properties_semver_neq(self):
+        prop = self.property(key="version", value="1.2.3", operator="semver_neq")
+        self.assertFalse(match_property(prop, {"version": "1.2.3"}))
+        self.assertTrue(match_property(prop, {"version": "1.2.4"}))
+        self.assertTrue(match_property(prop, {"version": "2.0.0"}))
+
+    def test_match_properties_semver_gt(self):
+        prop = self.property(key="version", value="1.2.3", operator="semver_gt")
+        self.assertTrue(match_property(prop, {"version": "1.2.4"}))
+        self.assertTrue(match_property(prop, {"version": "1.3.0"}))
+        self.assertTrue(match_property(prop, {"version": "2.0.0"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.3"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.2"}))
+        self.assertFalse(match_property(prop, {"version": "0.9.0"}))
+
+    def test_match_properties_semver_gte(self):
+        prop = self.property(key="version", value="1.2.3", operator="semver_gte")
+        self.assertTrue(match_property(prop, {"version": "1.2.3"}))
+        self.assertTrue(match_property(prop, {"version": "1.2.4"}))
+        self.assertTrue(match_property(prop, {"version": "2.0.0"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.2"}))
+        self.assertFalse(match_property(prop, {"version": "0.9.0"}))
+
+    def test_match_properties_semver_lt(self):
+        prop = self.property(key="version", value="1.2.3", operator="semver_lt")
+        self.assertTrue(match_property(prop, {"version": "1.2.2"}))
+        self.assertTrue(match_property(prop, {"version": "1.1.0"}))
+        self.assertTrue(match_property(prop, {"version": "0.9.0"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.3"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.4"}))
+        self.assertFalse(match_property(prop, {"version": "2.0.0"}))
+
+    def test_match_properties_semver_lte(self):
+        prop = self.property(key="version", value="1.2.3", operator="semver_lte")
+        self.assertTrue(match_property(prop, {"version": "1.2.3"}))
+        self.assertTrue(match_property(prop, {"version": "1.2.2"}))
+        self.assertTrue(match_property(prop, {"version": "0.9.0"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.4"}))
+        self.assertFalse(match_property(prop, {"version": "2.0.0"}))
+
+    def test_match_properties_semver_tilde(self):
+        # ~1.2.3 means >=1.2.3 <1.3.0
+        prop = self.property(key="version", value="1.2.3", operator="semver_tilde")
+        self.assertTrue(match_property(prop, {"version": "1.2.3"}))
+        self.assertTrue(match_property(prop, {"version": "1.2.5"}))
+        self.assertTrue(match_property(prop, {"version": "1.2.99"}))
+        self.assertFalse(match_property(prop, {"version": "1.3.0"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.2"}))
+        self.assertFalse(match_property(prop, {"version": "2.0.0"}))
+
+    def test_match_properties_semver_caret(self):
+        # ^1.2.3 means >=1.2.3 <2.0.0
+        prop = self.property(key="version", value="1.2.3", operator="semver_caret")
+        self.assertTrue(match_property(prop, {"version": "1.2.3"}))
+        self.assertTrue(match_property(prop, {"version": "1.9.0"}))
+        self.assertTrue(match_property(prop, {"version": "1.99.99"}))
+        self.assertFalse(match_property(prop, {"version": "2.0.0"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.2"}))
+        self.assertFalse(match_property(prop, {"version": "0.9.0"}))
+
+        # ^0.2.3 means >=0.2.3 <0.3.0 (leftmost non-zero is minor)
+        prop_zero_major = self.property(
+            key="version", value="0.2.3", operator="semver_caret"
+        )
+        self.assertTrue(match_property(prop_zero_major, {"version": "0.2.3"}))
+        self.assertTrue(match_property(prop_zero_major, {"version": "0.2.9"}))
+        self.assertFalse(match_property(prop_zero_major, {"version": "0.3.0"}))
+        self.assertFalse(match_property(prop_zero_major, {"version": "1.0.0"}))
+
+        # ^0.0.3 means >=0.0.3 <0.0.4 (leftmost non-zero is patch)
+        prop_zero_minor = self.property(
+            key="version", value="0.0.3", operator="semver_caret"
+        )
+        self.assertTrue(match_property(prop_zero_minor, {"version": "0.0.3"}))
+        self.assertFalse(match_property(prop_zero_minor, {"version": "0.0.4"}))
+        self.assertFalse(match_property(prop_zero_minor, {"version": "0.1.0"}))
+
+    def test_match_properties_semver_wildcard(self):
+        # 1.2.* means >=1.2.0 <1.3.0
+        prop = self.property(key="version", value="1.2.*", operator="semver_wildcard")
+        self.assertTrue(match_property(prop, {"version": "1.2.0"}))
+        self.assertTrue(match_property(prop, {"version": "1.2.5"}))
+        self.assertTrue(match_property(prop, {"version": "1.2.99"}))
+        self.assertFalse(match_property(prop, {"version": "1.3.0"}))
+        self.assertFalse(match_property(prop, {"version": "1.1.9"}))
+        self.assertFalse(match_property(prop, {"version": "2.0.0"}))
+
+        # 1.* means >=1.0.0 <2.0.0
+        prop_major = self.property(
+            key="version", value="1.*", operator="semver_wildcard"
+        )
+        self.assertTrue(match_property(prop_major, {"version": "1.0.0"}))
+        self.assertTrue(match_property(prop_major, {"version": "1.99.99"}))
+        self.assertFalse(match_property(prop_major, {"version": "2.0.0"}))
+        self.assertFalse(match_property(prop_major, {"version": "0.9.0"}))
+
+    def test_match_properties_semver_with_prerelease(self):
+        # Pre-release suffixes are stripped before comparison
+        prop = self.property(key="version", value="1.2.3", operator="semver_gt")
+        self.assertTrue(match_property(prop, {"version": "1.3.0-beta.1"}))
+        self.assertFalse(match_property(prop, {"version": "1.2.2-rc.1"}))
+
+        # Flag value can also have pre-release suffix
+        prop_pre = self.property(
+            key="version", value="1.2.3-alpha", operator="semver_gte"
+        )
+        self.assertTrue(match_property(prop_pre, {"version": "1.2.3"}))
+        self.assertTrue(match_property(prop_pre, {"version": "2.0.0"}))
+        self.assertFalse(match_property(prop_pre, {"version": "1.2.2"}))
+
+    def test_match_properties_semver_edge_cases(self):
+        """Test semver parsing handles v-prefix, whitespace, leading zeros, and other common formats."""
+        prop = self.property(key="version", value="1.2.3", operator="semver_eq")
+
+        # v-prefix: "v1.2.3" -> extracts "1.2.3"
+        self.assertTrue(match_property(prop, {"version": "v1.2.3"}))
+
+        # Leading space: " 1.2.3" -> extracts "1.2.3"
+        self.assertTrue(match_property(prop, {"version": " 1.2.3"}))
+
+        # Trailing space: "1.2.3 " -> extracts "1.2.3"
+        self.assertTrue(match_property(prop, {"version": "1.2.3 "}))
+
+        # Leading zeros: "01.02.03" -> int("01")=1, int("02")=2, int("03")=3
+        self.assertTrue(match_property(prop, {"version": "01.02.03"}))
+
+        # Flag value with v-prefix
+        prop_v = self.property(key="version", value="v1.2.3", operator="semver_eq")
+        self.assertTrue(match_property(prop_v, {"version": "1.2.3"}))
+
+        # 0.0.0 minimal version
+        prop_min = self.property(key="version", value="0.0.0", operator="semver_eq")
+        self.assertTrue(match_property(prop_min, {"version": "0.0.0"}))
+
+        prop_gt_min = self.property(key="version", value="0.0.0", operator="semver_gt")
+        self.assertTrue(match_property(prop_gt_min, {"version": "0.0.1"}))
+        self.assertFalse(match_property(prop_gt_min, {"version": "0.0.0"}))
+
+        # 4-part version: regex extracts "1.2.3.4" -> takes first 3 parts
+        prop_four = self.property(key="version", value="1.2.3", operator="semver_eq")
+        self.assertTrue(match_property(prop_four, {"version": "1.2.3.4"}))
+
+        # Truly invalid values raise InconclusiveMatchError
+        with self.assertRaises(InconclusiveMatchError):
+            match_property(prop, {"version": "abc"})
+
+        with self.assertRaises(InconclusiveMatchError):
+            match_property(prop, {"version": ""})
+
+        # Leading dot: ".1.2.3" -> invalid, empty first component
+        with self.assertRaises(InconclusiveMatchError):
+            match_property(prop, {"version": ".1.2.3"})
+
+        # Caret with v-prefix in flag value
+        prop_caret_v = self.property(
+            key="version", value="v1.2.3", operator="semver_caret"
+        )
+        self.assertTrue(match_property(prop_caret_v, {"version": "1.5.0"}))
+        self.assertFalse(match_property(prop_caret_v, {"version": "2.0.0"}))
+
+        # Wildcard with v-prefix in property value
+        prop_wild = self.property(
+            key="version", value="1.2.*", operator="semver_wildcard"
+        )
+        self.assertTrue(match_property(prop_wild, {"version": "v1.2.5"}))
+        self.assertFalse(match_property(prop_wild, {"version": "v1.3.0"}))
+
+    def test_match_properties_semver_invalid_values(self):
+        prop = self.property(key="version", value="1.2.3", operator="semver_eq")
+
+        # Invalid person property value
+        with self.assertRaises(InconclusiveMatchError):
+            match_property(prop, {"version": "not-a-version"})
+
+        # Missing key
+        with self.assertRaises(InconclusiveMatchError):
+            match_property(prop, {"other_key": "1.2.3"})
+
+        # None override value returns False (handled before semver logic)
+        self.assertFalse(match_property(prop, {"version": None}))
+
+        # Invalid flag value
+        prop_bad = self.property(key="version", value="not-valid", operator="semver_gt")
+        with self.assertRaises(InconclusiveMatchError):
+            match_property(prop_bad, {"version": "1.2.3"})
+
     def test_unknown_operator(self):
         property_a = self.property(key="key", value="2022-05-01", operator="is_unknown")
         with self.assertRaises(InconclusiveMatchError) as exception_context:
@@ -4266,13 +5026,15 @@ class TestCaptureCalls(unittest.TestCase):
             disable_geoip=False,
         )
 
-    @mock.patch("posthog.client.MAX_DICT_SIZE", 100)
     @mock.patch.object(Client, "capture")
     @mock.patch("posthog.client.flags")
     def test_capture_multiple_users_doesnt_out_of_memory(
         self, patch_flags, patch_capture
     ):
         client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+        # Set on the instance to avoid relying on module-constant patching behavior
+        # across Python/runtime implementations.
+        client.distinct_ids_feature_flags_reported.max_size = 100
         client.feature_flags = [
             {
                 "id": 1,

posthog/version.py

@@ -1,4 +1 @@
-VERSION = "7.6.0"
-
-if __name__ == "__main__":
-    print(VERSION, end="")  # noqa: T201
+VERSION = "7.9.7"

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "posthog"
-dynamic = ["version"]
+version = "7.9.7"
 description = "Integrate PostHog into any python application."
 authors = [{ name = "PostHog", email = "hey@posthog.com" }]
 maintainers = [{ name = "PostHog", email = "hey@posthog.com" }]
@@ -84,15 +84,15 @@ packages = [
     "posthog.ai",
     "posthog.ai.langchain",
     "posthog.ai.openai",
+    "posthog.ai.openai_agents",
     "posthog.ai.anthropic",
     "posthog.ai.gemini",
     "posthog.test",
+    "posthog.test.ai",
+    "posthog.test.ai.openai_agents",
     "posthog.integrations",
 ]
 
-[tool.setuptools.dynamic]
-version = { attr = "posthog.version.VERSION" }
-
 [tool.pytest.ini_options]
 asyncio_mode = "auto"
 asyncio_default_fixture_loop_scope = "function"

sdk_compliance_adapter/Dockerfile

@@ -0,0 +1,22 @@
+FROM python:3.12-slim
+
+WORKDIR /app
+
+# Copy the SDK source code
+COPY posthog/ /app/sdk/posthog/
+COPY setup.py pyproject.toml README.md LICENSE /app/sdk/
+
+# Install the SDK from source
+RUN cd /app/sdk && pip install --no-cache-dir -e .
+
+# Install adapter dependencies
+RUN pip install --no-cache-dir flask python-dateutil
+
+# Copy adapter code
+COPY sdk_compliance_adapter/adapter.py /app/adapter.py
+
+# Expose port 8080
+EXPOSE 8080
+
+# Run the adapter
+CMD ["python", "/app/adapter.py"]

sdk_compliance_adapter/README.md

@@ -0,0 +1,76 @@
+# PostHog Python SDK Test Adapter
+
+This adapter wraps the posthog-python SDK for compliance testing with the [PostHog SDK Test Harness](https://github.com/PostHog/posthog-sdk-test-harness).
+
+## What is This?
+
+This is a simple Flask app that:
+1. Wraps the posthog-python SDK
+2. Exposes a REST API for the test harness to control
+3. Tracks internal SDK state for test assertions
+
+## Running Tests
+
+Tests run automatically in CI via GitHub Actions. See the test harness repo for details.
+
+### Locally with Docker Compose
+
+```bash
+# From the posthog-python/sdk_compliance_adapter directory
+docker-compose up --build --abort-on-container-exit
+```
+
+This will:
+1. Build the Python SDK adapter
+2. Pull the test harness image
+3. Run all compliance tests
+4. Show results
+
+### Manually with Docker
+
+```bash
+# Create network
+docker network create test-network
+
+# Build and run adapter
+docker build -f sdk_compliance_adapter/Dockerfile -t posthog-python-adapter .
+docker run -d --name sdk-adapter --network test-network -p 8080:8080 posthog-python-adapter
+
+# Run test harness
+docker run --rm \
+  --name test-harness \
+  --network test-network \
+  ghcr.io/posthog/sdk-test-harness:latest \
+  run --adapter-url http://sdk-adapter:8080 --mock-url http://test-harness:8081
+
+# Cleanup
+docker stop sdk-adapter && docker rm sdk-adapter
+docker network rm test-network
+```
+
+## Adapter Implementation
+
+See [adapter.py](adapter.py) for the implementation.
+
+The adapter implements the standard SDK adapter interface defined in the [test harness CONTRACT](https://github.com/PostHog/posthog-sdk-test-harness/blob/main/CONTRACT.yaml):
+
+- `GET /health` - Return SDK information
+- `POST /init` - Initialize SDK with config
+- `POST /capture` - Capture an event
+- `POST /flush` - Flush pending events
+- `GET /state` - Return internal state
+- `POST /reset` - Reset SDK state
+
+### Key Implementation Details
+
+**Request Tracking**: The adapter monkey-patches `batch_post` to track all HTTP requests made by the SDK, including retries.
+
+**State Management**: Thread-safe state tracking for events captured vs sent, retry attempts, and errors.
+
+**UUID Tracking**: Extracts and tracks UUIDs from batches to verify deduplication.
+
+## Documentation
+
+For complete documentation on the test harness and how to implement adapters, see:
+- [PostHog SDK Test Harness](https://github.com/PostHog/posthog-sdk-test-harness)
+- [Adapter Implementation Guide](https://github.com/PostHog/posthog-sdk-test-harness/blob/main/ADAPTER_GUIDE.md)

sdk_compliance_adapter/adapter.py

@@ -0,0 +1,383 @@
+"""
+PostHog Python SDK Test Adapter
+
+This adapter implements the SDK Test Adapter Interface defined in the PostHog Capture API Contract.
+It wraps the posthog-python SDK and exposes a REST API for the test harness to exercise.
+"""
+
+import logging
+import os
+import threading
+import time
+from typing import Any, Dict, List, Optional
+
+from flask import Flask, jsonify, request
+
+from posthog import Client
+from posthog.request import batch_post as original_batch_post
+from posthog.version import VERSION
+
+# Configure logging
+logging.basicConfig(
+    level=logging.DEBUG, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
+)
+logger = logging.getLogger(__name__)
+
+app = Flask(__name__)
+
+
+class RequestInfo:
+    """Information about an HTTP request made by the SDK"""
+
+    def __init__(
+        self,
+        timestamp_ms: int,
+        status_code: int,
+        retry_attempt: int,
+        event_count: int,
+        uuid_list: List[str],
+    ):
+        self.timestamp_ms = timestamp_ms
+        self.status_code = status_code
+        self.retry_attempt = retry_attempt
+        self.event_count = event_count
+        self.uuid_list = uuid_list
+
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "timestamp_ms": self.timestamp_ms,
+            "status_code": self.status_code,
+            "retry_attempt": self.retry_attempt,
+            "event_count": self.event_count,
+            "uuid_list": self.uuid_list,
+        }
+
+
+class SDKState:
+    """Tracks SDK internal state for test assertions"""
+
+    def __init__(self):
+        self.lock = threading.Lock()
+        self.pending_events = 0
+        self.total_events_captured = 0
+        self.total_events_sent = 0
+        self.total_retries = 0
+        self.last_error: Optional[str] = None
+        self.requests_made: List[RequestInfo] = []
+        self.client: Optional[Client] = None
+        self.retry_attempts: Dict[str, int] = {}  # Track retry attempts by batch ID
+
+    def reset(self):
+        """Reset all state"""
+        with self.lock:
+            self.pending_events = 0
+            self.total_events_captured = 0
+            self.total_events_sent = 0
+            self.total_retries = 0
+            self.last_error = None
+            self.requests_made = []
+            self.retry_attempts = {}
+            if self.client:
+                # Flush and shutdown existing client
+                try:
+                    self.client.shutdown()
+                except Exception as e:
+                    logger.warning(f"Error shutting down client: {e}")
+                self.client = None
+
+    def increment_captured(self):
+        """Increment total events captured"""
+        with self.lock:
+            self.total_events_captured += 1
+            self.pending_events += 1
+
+    def record_request(self, status_code: int, batch: List[Dict], batch_id: str):
+        """Record an HTTP request made by the SDK"""
+        with self.lock:
+            # Determine retry attempt for this batch
+            retry_attempt = self.retry_attempts.get(batch_id, 0)
+
+            # Extract UUIDs from batch
+            uuid_list = [event.get("uuid", "") for event in batch]
+
+            request_info = RequestInfo(
+                timestamp_ms=int(time.time() * 1000),
+                status_code=status_code,
+                retry_attempt=retry_attempt,
+                event_count=len(batch),
+                uuid_list=uuid_list,
+            )
+            self.requests_made.append(request_info)
+
+            # Update counters
+            if status_code == 200:
+                # Success - clear pending events
+                self.total_events_sent += len(batch)
+                self.pending_events = max(0, self.pending_events - len(batch))
+                # Remove batch from retry tracking
+                self.retry_attempts.pop(batch_id, None)
+            else:
+                # Failure - increment retry count
+                self.retry_attempts[batch_id] = retry_attempt + 1
+                if retry_attempt > 0:
+                    self.total_retries += 1
+
+    def record_error(self, error: str):
+        """Record an error"""
+        with self.lock:
+            self.last_error = error
+
+    def get_state(self) -> Dict[str, Any]:
+        """Get current state as dict"""
+        with self.lock:
+            return {
+                "pending_events": self.pending_events,
+                "total_events_captured": self.total_events_captured,
+                "total_events_sent": self.total_events_sent,
+                "total_retries": self.total_retries,
+                "last_error": self.last_error,
+                "requests_made": [r.to_dict() for r in self.requests_made],
+            }
+
+
+# Global state
+state = SDKState()
+
+
+def create_batch_id(batch: List[Dict]) -> str:
+    """Create a unique ID for a batch based on UUIDs"""
+    uuids = sorted([event.get("uuid", "") for event in batch])
+    return "-".join(uuids[:3])  # Use first 3 UUIDs as batch ID
+
+
+def patched_batch_post(
+    api_key: str,
+    host: Optional[str] = None,
+    gzip: bool = False,
+    timeout: int = 15,
+    **kwargs,
+):
+    """Patched version of batch_post that tracks requests"""
+    batch = kwargs.get("batch", [])
+    batch_id = create_batch_id(batch)
+
+    try:
+        # Call original batch_post
+        response = original_batch_post(api_key, host, gzip, timeout, **kwargs)
+        # Record successful request
+        state.record_request(200, batch, batch_id)
+        return response
+    except Exception as e:
+        # Record failed request
+        status_code = (
+            getattr(e, "status_code", 500) if hasattr(e, "status_code") else 500
+        )
+        state.record_request(status_code, batch, batch_id)
+        state.record_error(str(e))
+        raise
+
+
+# Monkey-patch the batch_post function
+import posthog.request  # noqa: E402
+
+posthog.request.batch_post = patched_batch_post
+
+# Also patch in consumer module
+import posthog.consumer  # noqa: E402
+
+posthog.consumer.batch_post = patched_batch_post
+
+
+@app.route("/health", methods=["GET"])
+def health():
+    """Health check endpoint"""
+    return jsonify(
+        {
+            "sdk_name": "posthog-python",
+            "sdk_version": VERSION,
+            "adapter_version": "1.0.0",
+        }
+    )
+
+
+@app.route("/init", methods=["POST"])
+def init():
+    """Initialize the SDK client"""
+    try:
+        data = request.json or {}
+
+        # Reset state
+        state.reset()
+
+        # Extract config
+        api_key = data.get("api_key")
+        host = data.get("host")
+        flush_at = data.get("flush_at", 100)
+        flush_interval_ms = data.get("flush_interval_ms", 500)
+        max_retries = data.get("max_retries", 3)
+        enable_compression = data.get("enable_compression", False)
+
+        if not api_key:
+            return jsonify({"error": "api_key is required"}), 400
+        if not host:
+            return jsonify({"error": "host is required"}), 400
+
+        # Convert flush_interval from ms to seconds
+        flush_interval = flush_interval_ms / 1000.0
+
+        # Create client
+        client = Client(
+            project_api_key=api_key,
+            host=host,
+            flush_at=flush_at,
+            flush_interval=flush_interval,
+            gzip=enable_compression,
+            max_retries=max_retries,
+            debug=True,
+        )
+
+        state.client = client
+
+        logger.info(
+            f"Initialized SDK with api_key={api_key[:10]}..., host={host}, "
+            f"flush_at={flush_at}, flush_interval={flush_interval}, "
+            f"max_retries={max_retries}, gzip={enable_compression}"
+        )
+
+        return jsonify({"success": True})
+    except Exception as e:
+        logger.exception("Error initializing SDK")
+        return jsonify({"error": str(e)}), 500
+
+
+@app.route("/capture", methods=["POST"])
+def capture():
+    """Capture a single event"""
+    try:
+        if not state.client:
+            return jsonify({"error": "SDK not initialized"}), 400
+
+        data = request.json or {}
+
+        distinct_id = data.get("distinct_id")
+        event = data.get("event")
+        properties = data.get("properties")
+        timestamp = data.get("timestamp")
+
+        if not distinct_id:
+            return jsonify({"error": "distinct_id is required"}), 400
+        if not event:
+            return jsonify({"error": "event is required"}), 400
+
+        # Capture event
+        kwargs = {"distinct_id": distinct_id, "properties": properties}
+        if timestamp:
+            # Parse ISO8601 timestamp
+            from dateutil.parser import parse  # type: ignore[import-untyped]
+
+            kwargs["timestamp"] = parse(timestamp)
+
+        uuid = state.client.capture(event, **kwargs)
+
+        # Track that we captured an event
+        state.increment_captured()
+
+        logger.info(f"Captured event: {event} for {distinct_id}, uuid={uuid}")
+
+        return jsonify({"success": True, "uuid": uuid})
+    except Exception as e:
+        logger.exception("Error capturing event")
+        state.record_error(str(e))
+        return jsonify({"error": str(e)}), 500
+
+
+@app.route("/identify", methods=["POST"])
+def identify():
+    """Identify a user"""
+    try:
+        if not state.client:
+            return jsonify({"error": "SDK not initialized"}), 400
+
+        data = request.json or {}
+
+        distinct_id = data.get("distinct_id")
+        properties = data.get("properties")
+        properties_set_once = data.get("properties_set_once")
+
+        if not distinct_id:
+            return jsonify({"error": "distinct_id is required"}), 400
+
+        # Use the identify pattern - set + set_once
+        if properties:
+            state.client.set(distinct_id=distinct_id, properties=properties)
+            state.increment_captured()
+
+        if properties_set_once:
+            state.client.set_once(
+                distinct_id=distinct_id, properties=properties_set_once
+            )
+            state.increment_captured()
+
+        logger.info(f"Identified user: {distinct_id}")
+
+        return jsonify({"success": True})
+    except Exception as e:
+        logger.exception("Error identifying user")
+        state.record_error(str(e))
+        return jsonify({"error": str(e)}), 500
+
+
+@app.route("/flush", methods=["POST"])
+def flush():
+    """Force flush all pending events"""
+    try:
+        if not state.client:
+            return jsonify({"error": "SDK not initialized"}), 400
+
+        # Flush and wait
+        state.client.flush()
+
+        # Wait a bit for flush to complete
+        # The flush() method triggers queue.join() which blocks until all items are processed
+        time.sleep(0.5)
+
+        logger.info("Flushed pending events")
+
+        return jsonify({"success": True, "events_flushed": state.total_events_sent})
+    except Exception as e:
+        logger.exception("Error flushing events")
+        state.record_error(str(e))
+        return jsonify({"error": str(e), "errors": [str(e)]}, 500)
+
+
+@app.route("/state", methods=["GET"])
+def get_state():
+    """Get internal SDK state"""
+    try:
+        return jsonify(state.get_state())
+    except Exception as e:
+        logger.exception("Error getting state")
+        return jsonify({"error": str(e)}), 500
+
+
+@app.route("/reset", methods=["POST"])
+def reset():
+    """Reset SDK state"""
+    try:
+        state.reset()
+        logger.info("Reset SDK state")
+        return jsonify({"success": True})
+    except Exception as e:
+        logger.exception("Error resetting state")
+        return jsonify({"error": str(e)}), 500
+
+
+def main():
+    """Main entry point"""
+    port = int(os.environ.get("PORT", 8080))
+    logger.info(f"Starting SDK Test Adapter on port {port}")
+    app.run(host="0.0.0.0", port=port, debug=False)
+
+
+if __name__ == "__main__":
+    main()

sdk_compliance_adapter/docker-compose.yml

@@ -0,0 +1,25 @@
+version: "3.8"
+
+services:
+  # PostHog Python SDK adapter
+  sdk-adapter:
+    build:
+      context: ..
+      dockerfile: sdk_compliance_adapter/Dockerfile
+    ports:
+      - "8080:8080"
+    networks:
+      - test-network
+
+  # Test harness
+  test-harness:
+    image: ghcr.io/posthog/sdk-test-harness:latest
+    command: ["run", "--adapter-url", "http://sdk-adapter:8080", "--mock-url", "http://test-harness:8081"]
+    networks:
+      - test-network
+    depends_on:
+      - sdk-adapter
+
+networks:
+  test-network:
+    driver: bridge

sdk_compliance_adapter/requirements.txt

@@ -0,0 +1,3 @@
+# SDK Test Adapter dependencies
+flask>=3.0.0
+python-dateutil>=2.8.0

setup_analytics.py

@@ -23,9 +23,7 @@ with open("pyproject.toml", "rb") as f:
 
 # Override specific values
 config["project"]["name"] = "posthoganalytics"
-config["tool"]["setuptools"]["dynamic"]["version"] = {
-    "attr": "posthoganalytics.version.VERSION"
-}
+config["project"]["readme"] = "README_ANALYTICS.md"
 
 # Rename packages from posthog.* to posthoganalytics.*
 if "packages" in config["tool"]["setuptools"]:

uv.lock

@@ -278,43 +278,31 @@ sdist = { url = "https://files.pythonhosted.org/packages/eb/56/b1ba7935a17738ae8
 wheels = [
     { url = "https://files.pythonhosted.org/packages/50/bd/b1a6362b80628111e6653c961f987faa55262b4002fcec42308cad1db680/cffi-2.0.0-cp310-cp310-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:53f77cbe57044e88bbd5ed26ac1d0514d2acf0591dd6bb02a3ae37f76811b80c", size = 208811, upload-time = "2025-09-08T23:22:12.267Z" },
     { url = "https://files.pythonhosted.org/packages/4f/27/6933a8b2562d7bd1fb595074cf99cc81fc3789f6a6c05cdabb46284a3188/cffi-2.0.0-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:3e837e369566884707ddaf85fc1744b47575005c0a229de3327f8f9a20f4efeb", size = 216402, upload-time = "2025-09-08T23:22:13.455Z" },
-    { url = "https://files.pythonhosted.org/packages/05/eb/b86f2a2645b62adcfff53b0dd97e8dfafb5c8aa864bd0d9a2c2049a0d551/cffi-2.0.0-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:5eda85d6d1879e692d546a078b44251cdd08dd1cfb98dfb77b670c97cee49ea0", size = 203217, upload-time = "2025-09-08T23:22:14.596Z" },
-    { url = "https://files.pythonhosted.org/packages/9f/e0/6cbe77a53acf5acc7c08cc186c9928864bd7c005f9efd0d126884858a5fe/cffi-2.0.0-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:9332088d75dc3241c702d852d4671613136d90fa6881da7d770a483fd05248b4", size = 203079, upload-time = "2025-09-08T23:22:15.769Z" },
     { url = "https://files.pythonhosted.org/packages/98/29/9b366e70e243eb3d14a5cb488dfd3a0b6b2f1fb001a203f653b93ccfac88/cffi-2.0.0-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:fc7de24befaeae77ba923797c7c87834c73648a05a4bde34b3b7e5588973a453", size = 216475, upload-time = "2025-09-08T23:22:17.427Z" },
     { url = "https://files.pythonhosted.org/packages/21/7a/13b24e70d2f90a322f2900c5d8e1f14fa7e2a6b3332b7309ba7b2ba51a5a/cffi-2.0.0-cp310-cp310-musllinux_1_2_aarch64.whl", hash = "sha256:cf364028c016c03078a23b503f02058f1814320a56ad535686f90565636a9495", size = 218829, upload-time = "2025-09-08T23:22:19.069Z" },
     { url = "https://files.pythonhosted.org/packages/60/99/c9dc110974c59cc981b1f5b66e1d8af8af764e00f0293266824d9c4254bc/cffi-2.0.0-cp310-cp310-musllinux_1_2_i686.whl", hash = "sha256:e11e82b744887154b182fd3e7e8512418446501191994dbf9c9fc1f32cc8efd5", size = 211211, upload-time = "2025-09-08T23:22:20.588Z" },
     { url = "https://files.pythonhosted.org/packages/49/72/ff2d12dbf21aca1b32a40ed792ee6b40f6dc3a9cf1644bd7ef6e95e0ac5e/cffi-2.0.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:8ea985900c5c95ce9db1745f7933eeef5d314f0565b27625d9a10ec9881e1bfb", size = 218036, upload-time = "2025-09-08T23:22:22.143Z" },
     { url = "https://files.pythonhosted.org/packages/b1/b7/1200d354378ef52ec227395d95c2576330fd22a869f7a70e88e1447eb234/cffi-2.0.0-cp311-cp311-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:baf5215e0ab74c16e2dd324e8ec067ef59e41125d3eade2b863d294fd5035c92", size = 209613, upload-time = "2025-09-08T23:22:29.475Z" },
     { url = "https://files.pythonhosted.org/packages/b8/56/6033f5e86e8cc9bb629f0077ba71679508bdf54a9a5e112a3c0b91870332/cffi-2.0.0-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:730cacb21e1bdff3ce90babf007d0a0917cc3e6492f336c2f0134101e0944f93", size = 216476, upload-time = "2025-09-08T23:22:31.063Z" },
-    { url = "https://files.pythonhosted.org/packages/dc/7f/55fecd70f7ece178db2f26128ec41430d8720f2d12ca97bf8f0a628207d5/cffi-2.0.0-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:6824f87845e3396029f3820c206e459ccc91760e8fa24422f8b0c3d1731cbec5", size = 203374, upload-time = "2025-09-08T23:22:32.507Z" },
-    { url = "https://files.pythonhosted.org/packages/84/ef/a7b77c8bdc0f77adc3b46888f1ad54be8f3b7821697a7b89126e829e676a/cffi-2.0.0-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:9de40a7b0323d889cf8d23d1ef214f565ab154443c42737dfe52ff82cf857664", size = 202597, upload-time = "2025-09-08T23:22:34.132Z" },
     { url = "https://files.pythonhosted.org/packages/d7/91/500d892b2bf36529a75b77958edfcd5ad8e2ce4064ce2ecfeab2125d72d1/cffi-2.0.0-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:8941aaadaf67246224cee8c3803777eed332a19d909b47e29c9842ef1e79ac26", size = 215574, upload-time = "2025-09-08T23:22:35.443Z" },
     { url = "https://files.pythonhosted.org/packages/44/64/58f6255b62b101093d5df22dcb752596066c7e89dd725e0afaed242a61be/cffi-2.0.0-cp311-cp311-musllinux_1_2_aarch64.whl", hash = "sha256:a05d0c237b3349096d3981b727493e22147f934b20f6f125a3eba8f994bec4a9", size = 218971, upload-time = "2025-09-08T23:22:36.805Z" },
     { url = "https://files.pythonhosted.org/packages/ab/49/fa72cebe2fd8a55fbe14956f9970fe8eb1ac59e5df042f603ef7c8ba0adc/cffi-2.0.0-cp311-cp311-musllinux_1_2_i686.whl", hash = "sha256:94698a9c5f91f9d138526b48fe26a199609544591f859c870d477351dc7b2414", size = 211972, upload-time = "2025-09-08T23:22:38.436Z" },
     { url = "https://files.pythonhosted.org/packages/0b/28/dd0967a76aab36731b6ebfe64dec4e981aff7e0608f60c2d46b46982607d/cffi-2.0.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:5fed36fccc0612a53f1d4d9a816b50a36702c28a2aa880cb8a122b3466638743", size = 217078, upload-time = "2025-09-08T23:22:39.776Z" },
     { url = "https://files.pythonhosted.org/packages/ff/df/a4f0fbd47331ceeba3d37c2e51e9dfc9722498becbeec2bd8bc856c9538a/cffi-2.0.0-cp312-cp312-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:21d1152871b019407d8ac3985f6775c079416c282e431a4da6afe7aefd2bccbe", size = 212529, upload-time = "2025-09-08T23:22:47.349Z" },
     { url = "https://files.pythonhosted.org/packages/d5/72/12b5f8d3865bf0f87cf1404d8c374e7487dcf097a1c91c436e72e6badd83/cffi-2.0.0-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:b21e08af67b8a103c71a250401c78d5e0893beff75e28c53c98f4de42f774062", size = 220097, upload-time = "2025-09-08T23:22:48.677Z" },
-    { url = "https://files.pythonhosted.org/packages/c2/95/7a135d52a50dfa7c882ab0ac17e8dc11cec9d55d2c18dda414c051c5e69e/cffi-2.0.0-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:1e3a615586f05fc4065a8b22b8152f0c1b00cdbc60596d187c2a74f9e3036e4e", size = 207983, upload-time = "2025-09-08T23:22:50.06Z" },
-    { url = "https://files.pythonhosted.org/packages/3a/c8/15cb9ada8895957ea171c62dc78ff3e99159ee7adb13c0123c001a2546c1/cffi-2.0.0-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:81afed14892743bbe14dacb9e36d9e0e504cd204e0b165062c488942b9718037", size = 206519, upload-time = "2025-09-08T23:22:51.364Z" },
     { url = "https://files.pythonhosted.org/packages/78/2d/7fa73dfa841b5ac06c7b8855cfc18622132e365f5b81d02230333ff26e9e/cffi-2.0.0-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:3e17ed538242334bf70832644a32a7aae3d83b57567f9fd60a26257e992b79ba", size = 219572, upload-time = "2025-09-08T23:22:52.902Z" },
     { url = "https://files.pythonhosted.org/packages/07/e0/267e57e387b4ca276b90f0434ff88b2c2241ad72b16d31836adddfd6031b/cffi-2.0.0-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:3925dd22fa2b7699ed2617149842d2e6adde22b262fcbfada50e3d195e4b3a94", size = 222963, upload-time = "2025-09-08T23:22:54.518Z" },
     { url = "https://files.pythonhosted.org/packages/b6/75/1f2747525e06f53efbd878f4d03bac5b859cbc11c633d0fb81432d98a795/cffi-2.0.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:2c8f814d84194c9ea681642fd164267891702542f028a15fc97d4674b6206187", size = 221361, upload-time = "2025-09-08T23:22:55.867Z" },
     { url = "https://files.pythonhosted.org/packages/b0/1e/d22cc63332bd59b06481ceaac49d6c507598642e2230f201649058a7e704/cffi-2.0.0-cp313-cp313-manylinux1_i686.manylinux2014_i686.manylinux_2_17_i686.manylinux_2_5_i686.whl", hash = "sha256:07b271772c100085dd28b74fa0cd81c8fb1a3ba18b21e03d7c27f3436a10606b", size = 212446, upload-time = "2025-09-08T23:23:03.472Z" },
     { url = "https://files.pythonhosted.org/packages/a9/f5/a2c23eb03b61a0b8747f211eb716446c826ad66818ddc7810cc2cc19b3f2/cffi-2.0.0-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:d48a880098c96020b02d5a1f7d9251308510ce8858940e6fa99ece33f610838b", size = 220101, upload-time = "2025-09-08T23:23:04.792Z" },
-    { url = "https://files.pythonhosted.org/packages/f2/7f/e6647792fc5850d634695bc0e6ab4111ae88e89981d35ac269956605feba/cffi-2.0.0-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:f93fd8e5c8c0a4aa1f424d6173f14a892044054871c771f8566e4008eaa359d2", size = 207948, upload-time = "2025-09-08T23:23:06.127Z" },
-    { url = "https://files.pythonhosted.org/packages/cb/1e/a5a1bd6f1fb30f22573f76533de12a00bf274abcdc55c8edab639078abb6/cffi-2.0.0-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:dd4f05f54a52fb558f1ba9f528228066954fee3ebe629fc1660d874d040ae5a3", size = 206422, upload-time = "2025-09-08T23:23:07.753Z" },
     { url = "https://files.pythonhosted.org/packages/98/df/0a1755e750013a2081e863e7cd37e0cdd02664372c754e5560099eb7aa44/cffi-2.0.0-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:c8d3b5532fc71b7a77c09192b4a5a200ea992702734a2e9279a37f2478236f26", size = 219499, upload-time = "2025-09-08T23:23:09.648Z" },
     { url = "https://files.pythonhosted.org/packages/50/e1/a969e687fcf9ea58e6e2a928ad5e2dd88cc12f6f0ab477e9971f2309b57c/cffi-2.0.0-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:d9b29c1f0ae438d5ee9acb31cadee00a58c46cc9c0b2f9038c6b0b3470877a8c", size = 222928, upload-time = "2025-09-08T23:23:10.928Z" },
     { url = "https://files.pythonhosted.org/packages/36/54/0362578dd2c9e557a28ac77698ed67323ed5b9775ca9d3fe73fe191bb5d8/cffi-2.0.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:6d50360be4546678fc1b79ffe7a66265e28667840010348dd69a314145807a1b", size = 221302, upload-time = "2025-09-08T23:23:12.42Z" },
     { url = "https://files.pythonhosted.org/packages/d6/43/0e822876f87ea8a4ef95442c3d766a06a51fc5298823f884ef87aaad168c/cffi-2.0.0-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:24b6f81f1983e6df8db3adc38562c83f7d4a0c36162885ec7f7b77c7dcbec97b", size = 220049, upload-time = "2025-09-08T23:23:20.853Z" },
-    { url = "https://files.pythonhosted.org/packages/b4/89/76799151d9c2d2d1ead63c2429da9ea9d7aac304603de0c6e8764e6e8e70/cffi-2.0.0-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:12873ca6cb9b0f0d3a0da705d6086fe911591737a59f28b7936bdfed27c0d47c", size = 207793, upload-time = "2025-09-08T23:23:22.08Z" },
-    { url = "https://files.pythonhosted.org/packages/bb/dd/3465b14bb9e24ee24cb88c9e3730f6de63111fffe513492bf8c808a3547e/cffi-2.0.0-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:d9b97165e8aed9272a6bb17c01e3cc5871a594a446ebedc996e2397a1c1ea8ef", size = 206300, upload-time = "2025-09-08T23:23:23.314Z" },
     { url = "https://files.pythonhosted.org/packages/47/d9/d83e293854571c877a92da46fdec39158f8d7e68da75bf73581225d28e90/cffi-2.0.0-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:afb8db5439b81cf9c9d0c80404b60c3cc9c3add93e114dcae767f1477cb53775", size = 219244, upload-time = "2025-09-08T23:23:24.541Z" },
     { url = "https://files.pythonhosted.org/packages/2b/0f/1f177e3683aead2bb00f7679a16451d302c436b5cbf2505f0ea8146ef59e/cffi-2.0.0-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:737fe7d37e1a1bffe70bd5754ea763a62a066dc5913ca57e957824b72a85e205", size = 222828, upload-time = "2025-09-08T23:23:26.143Z" },
     { url = "https://files.pythonhosted.org/packages/c6/0f/cafacebd4b040e3119dcb32fed8bdef8dfe94da653155f9d0b9dc660166e/cffi-2.0.0-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:38100abb9d1b1435bc4cc340bb4489635dc2f0da7456590877030c9b3d40b0c1", size = 220926, upload-time = "2025-09-08T23:23:27.873Z" },
     { url = "https://files.pythonhosted.org/packages/be/b4/c56878d0d1755cf9caa54ba71e5d049479c52f9e4afc230f06822162ab2f/cffi-2.0.0-cp314-cp314t-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:7cc09976e8b56f8cebd752f7113ad07752461f48a58cbba644139015ac24954c", size = 221593, upload-time = "2025-09-08T23:23:31.91Z" },
-    { url = "https://files.pythonhosted.org/packages/e0/0d/eb704606dfe8033e7128df5e90fee946bbcb64a04fcdaa97321309004000/cffi-2.0.0-cp314-cp314t-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:92b68146a71df78564e4ef48af17551a5ddd142e5190cdf2c5624d0c3ff5b2e8", size = 209354, upload-time = "2025-09-08T23:23:33.214Z" },
-    { url = "https://files.pythonhosted.org/packages/d8/19/3c435d727b368ca475fb8742ab97c9cb13a0de600ce86f62eab7fa3eea60/cffi-2.0.0-cp314-cp314t-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:b1e74d11748e7e98e2f426ab176d4ed720a64412b6a15054378afdb71e0f37dc", size = 208480, upload-time = "2025-09-08T23:23:34.495Z" },
     { url = "https://files.pythonhosted.org/packages/d0/44/681604464ed9541673e486521497406fadcc15b5217c3e326b061696899a/cffi-2.0.0-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:28a3a209b96630bca57cce802da70c266eb08c6e97e5afd61a75611ee6c64592", size = 221584, upload-time = "2025-09-08T23:23:36.096Z" },
     { url = "https://files.pythonhosted.org/packages/25/8e/342a504ff018a2825d395d44d63a767dd8ebc927ebda557fecdaca3ac33a/cffi-2.0.0-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:7553fb2090d71822f02c629afe6042c299edf91ba1bf94951165613553984512", size = 224443, upload-time = "2025-09-08T23:23:37.328Z" },
     { url = "https://files.pythonhosted.org/packages/e1/5e/b666bacbbc60fbf415ba9988324a132c9a7a0448a9a8f125074671c0f2c3/cffi-2.0.0-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:6c6c373cfc5c83a975506110d17457138c8c63016b563cc9ed6e056a82f13ce4", size = 223437, upload-time = "2025-09-08T23:23:38.945Z" },
@@ -540,10 +528,8 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/5c/49/498c86566a1d80e978b42f0d702795f69887005548c041636df6ae1ca64c/cryptography-46.0.3-cp311-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:01ca9ff2885f3acc98c29f1860552e37f6d7c7d013d7334ff2a9de43a449315d", size = 4450807, upload-time = "2025-10-15T23:16:56.414Z" },
     { url = "https://files.pythonhosted.org/packages/4b/0a/863a3604112174c8624a2ac3c038662d9e59970c7f926acdcfaed8d61142/cryptography-46.0.3-cp311-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:6eae65d4c3d33da080cff9c4ab1f711b15c1d9760809dad6ea763f3812d254cb", size = 4299615, upload-time = "2025-10-15T23:16:58.442Z" },
     { url = "https://files.pythonhosted.org/packages/64/02/b73a533f6b64a69f3cd3872acb6ebc12aef924d8d103133bb3ea750dc703/cryptography-46.0.3-cp311-abi3-manylinux_2_28_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:e5bf0ed4490068a2e72ac03d786693adeb909981cc596425d09032d372bcc849", size = 4016800, upload-time = "2025-10-15T23:17:00.378Z" },
-    { url = "https://files.pythonhosted.org/packages/25/d5/16e41afbfa450cde85a3b7ec599bebefaef16b5c6ba4ec49a3532336ed72/cryptography-46.0.3-cp311-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:5ecfccd2329e37e9b7112a888e76d9feca2347f12f37918facbb893d7bb88ee8", size = 4984707, upload-time = "2025-10-15T23:17:01.98Z" },
     { url = "https://files.pythonhosted.org/packages/c9/56/e7e69b427c3878352c2fb9b450bd0e19ed552753491d39d7d0a2f5226d41/cryptography-46.0.3-cp311-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:a2c0cd47381a3229c403062f764160d57d4d175e022c1df84e168c6251a22eec", size = 4482541, upload-time = "2025-10-15T23:17:04.078Z" },
     { url = "https://files.pythonhosted.org/packages/78/f6/50736d40d97e8483172f1bb6e698895b92a223dba513b0ca6f06b2365339/cryptography-46.0.3-cp311-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:549e234ff32571b1f4076ac269fcce7a808d3bf98b76c8dd560e42dbc66d7d91", size = 4299464, upload-time = "2025-10-15T23:17:05.483Z" },
-    { url = "https://files.pythonhosted.org/packages/00/de/d8e26b1a855f19d9994a19c702fa2e93b0456beccbcfe437eda00e0701f2/cryptography-46.0.3-cp311-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:c0a7bb1a68a5d3471880e264621346c48665b3bf1c3759d682fc0864c540bd9e", size = 4950838, upload-time = "2025-10-15T23:17:07.425Z" },
     { url = "https://files.pythonhosted.org/packages/8f/29/798fc4ec461a1c9e9f735f2fc58741b0daae30688f41b2497dcbc9ed1355/cryptography-46.0.3-cp311-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:10b01676fc208c3e6feeb25a8b83d81767e8059e1fe86e1dc62d10a3018fa926", size = 4481596, upload-time = "2025-10-15T23:17:09.343Z" },
     { url = "https://files.pythonhosted.org/packages/15/8d/03cd48b20a573adfff7652b76271078e3045b9f49387920e7f1f631d125e/cryptography-46.0.3-cp311-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:0abf1ffd6e57c67e92af68330d05760b7b7efb243aab8377e583284dbab72c71", size = 4426782, upload-time = "2025-10-15T23:17:11.22Z" },
     { url = "https://files.pythonhosted.org/packages/fa/b1/ebacbfe53317d55cf33165bda24c86523497a6881f339f9aae5c2e13e57b/cryptography-46.0.3-cp311-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:a04bee9ab6a4da801eb9b51f1b708a1b5b5c9eb48c03f74198464c66f0d344ac", size = 4698381, upload-time = "2025-10-15T23:17:12.829Z" },
@@ -551,10 +537,8 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/c5/fd/bc1daf8230eaa075184cbbf5f8cd00ba9db4fd32d63fb83da4671b72ed8a/cryptography-46.0.3-cp314-cp314t-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:39b6755623145ad5eff1dab323f4eae2a32a77a7abef2c5089a04a3d04366715", size = 4435078, upload-time = "2025-10-15T23:17:23.042Z" },
     { url = "https://files.pythonhosted.org/packages/82/98/d3bd5407ce4c60017f8ff9e63ffee4200ab3e23fe05b765cab805a7db008/cryptography-46.0.3-cp314-cp314t-manylinux_2_28_aarch64.whl", hash = "sha256:db391fa7c66df6762ee3f00c95a89e6d428f4d60e7abc8328f4fe155b5ac6e54", size = 4293460, upload-time = "2025-10-15T23:17:24.885Z" },
     { url = "https://files.pythonhosted.org/packages/26/e9/e23e7900983c2b8af7a08098db406cf989d7f09caea7897e347598d4cd5b/cryptography-46.0.3-cp314-cp314t-manylinux_2_28_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:78a97cf6a8839a48c49271cdcbd5cf37ca2c1d6b7fdd86cc864f302b5e9bf459", size = 3995237, upload-time = "2025-10-15T23:17:26.449Z" },
-    { url = "https://files.pythonhosted.org/packages/91/15/af68c509d4a138cfe299d0d7ddb14afba15233223ebd933b4bbdbc7155d3/cryptography-46.0.3-cp314-cp314t-manylinux_2_28_ppc64le.whl", hash = "sha256:dfb781ff7eaa91a6f7fd41776ec37c5853c795d3b358d4896fdbb5df168af422", size = 4967344, upload-time = "2025-10-15T23:17:28.06Z" },
     { url = "https://files.pythonhosted.org/packages/ca/e3/8643d077c53868b681af077edf6b3cb58288b5423610f21c62aadcbe99f4/cryptography-46.0.3-cp314-cp314t-manylinux_2_28_x86_64.whl", hash = "sha256:6f61efb26e76c45c4a227835ddeae96d83624fb0d29eb5df5b96e14ed1a0afb7", size = 4466564, upload-time = "2025-10-15T23:17:29.665Z" },
     { url = "https://files.pythonhosted.org/packages/0e/43/c1e8726fa59c236ff477ff2b5dc071e54b21e5a1e51aa2cee1676f1c986f/cryptography-46.0.3-cp314-cp314t-manylinux_2_34_aarch64.whl", hash = "sha256:23b1a8f26e43f47ceb6d6a43115f33a5a37d57df4ea0ca295b780ae8546e8044", size = 4292415, upload-time = "2025-10-15T23:17:31.686Z" },
-    { url = "https://files.pythonhosted.org/packages/42/f9/2f8fefdb1aee8a8e3256a0568cffc4e6d517b256a2fe97a029b3f1b9fe7e/cryptography-46.0.3-cp314-cp314t-manylinux_2_34_ppc64le.whl", hash = "sha256:b419ae593c86b87014b9be7396b385491ad7f320bde96826d0dd174459e54665", size = 4931457, upload-time = "2025-10-15T23:17:33.478Z" },
     { url = "https://files.pythonhosted.org/packages/79/30/9b54127a9a778ccd6d27c3da7563e9f2d341826075ceab89ae3b41bf5be2/cryptography-46.0.3-cp314-cp314t-manylinux_2_34_x86_64.whl", hash = "sha256:50fc3343ac490c6b08c0cf0d704e881d0d660be923fd3076db3e932007e726e3", size = 4466074, upload-time = "2025-10-15T23:17:35.158Z" },
     { url = "https://files.pythonhosted.org/packages/ac/68/b4f4a10928e26c941b1b6a179143af9f4d27d88fe84a6a3c53592d2e76bf/cryptography-46.0.3-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:22d7e97932f511d6b0b04f2bfd818d73dcd5928db509460aaf48384778eb6d20", size = 4420569, upload-time = "2025-10-15T23:17:37.188Z" },
     { url = "https://files.pythonhosted.org/packages/a3/49/3746dab4c0d1979888f125226357d3262a6dd40e114ac29e3d2abdf1ec55/cryptography-46.0.3-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:d55f3dffadd674514ad19451161118fd010988540cee43d8bc20675e775925de", size = 4681941, upload-time = "2025-10-15T23:17:39.236Z" },
@@ -562,10 +546,8 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/26/42/fa8389d4478368743e24e61eea78846a0006caffaf72ea24a15159215a14/cryptography-46.0.3-cp38-abi3-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:15ab9b093e8f09daab0f2159bb7e47532596075139dd74365da52ecc9cb46c5d", size = 4440029, upload-time = "2025-10-15T23:17:49.837Z" },
     { url = "https://files.pythonhosted.org/packages/5f/eb/f483db0ec5ac040824f269e93dd2bd8a21ecd1027e77ad7bdf6914f2fd80/cryptography-46.0.3-cp38-abi3-manylinux_2_28_aarch64.whl", hash = "sha256:46acf53b40ea38f9c6c229599a4a13f0d46a6c3fa9ef19fc1a124d62e338dfa0", size = 4297222, upload-time = "2025-10-15T23:17:51.357Z" },
     { url = "https://files.pythonhosted.org/packages/fd/cf/da9502c4e1912cb1da3807ea3618a6829bee8207456fbbeebc361ec38ba3/cryptography-46.0.3-cp38-abi3-manylinux_2_28_armv7l.manylinux_2_31_armv7l.whl", hash = "sha256:10ca84c4668d066a9878890047f03546f3ae0a6b8b39b697457b7757aaf18dbc", size = 4012280, upload-time = "2025-10-15T23:17:52.964Z" },
-    { url = "https://files.pythonhosted.org/packages/6b/8f/9adb86b93330e0df8b3dcf03eae67c33ba89958fc2e03862ef1ac2b42465/cryptography-46.0.3-cp38-abi3-manylinux_2_28_ppc64le.whl", hash = "sha256:36e627112085bb3b81b19fed209c05ce2a52ee8b15d161b7c643a7d5a88491f3", size = 4978958, upload-time = "2025-10-15T23:17:54.965Z" },
     { url = "https://files.pythonhosted.org/packages/d1/a0/5fa77988289c34bdb9f913f5606ecc9ada1adb5ae870bd0d1054a7021cc4/cryptography-46.0.3-cp38-abi3-manylinux_2_28_x86_64.whl", hash = "sha256:1000713389b75c449a6e979ffc7dcc8ac90b437048766cef052d4d30b8220971", size = 4473714, upload-time = "2025-10-15T23:17:56.754Z" },
     { url = "https://files.pythonhosted.org/packages/14/e5/fc82d72a58d41c393697aa18c9abe5ae1214ff6f2a5c18ac470f92777895/cryptography-46.0.3-cp38-abi3-manylinux_2_34_aarch64.whl", hash = "sha256:b02cf04496f6576afffef5ddd04a0cb7d49cf6be16a9059d793a30b035f6b6ac", size = 4296970, upload-time = "2025-10-15T23:17:58.588Z" },
-    { url = "https://files.pythonhosted.org/packages/78/06/5663ed35438d0b09056973994f1aec467492b33bd31da36e468b01ec1097/cryptography-46.0.3-cp38-abi3-manylinux_2_34_ppc64le.whl", hash = "sha256:71e842ec9bc7abf543b47cf86b9a743baa95f4677d22baa4c7d5c69e49e9bc04", size = 4940236, upload-time = "2025-10-15T23:18:00.897Z" },
     { url = "https://files.pythonhosted.org/packages/fc/59/873633f3f2dcd8a053b8dd1d38f783043b5fce589c0f6988bf55ef57e43e/cryptography-46.0.3-cp38-abi3-manylinux_2_34_x86_64.whl", hash = "sha256:402b58fc32614f00980b66d6e56a5b4118e6cb362ae8f3fda141ba4689bd4506", size = 4472642, upload-time = "2025-10-15T23:18:02.749Z" },
     { url = "https://files.pythonhosted.org/packages/3d/39/8e71f3930e40f6877737d6f69248cf74d4e34b886a3967d32f919cc50d3b/cryptography-46.0.3-cp38-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:ef639cb3372f69ec44915fafcd6698b6cc78fbe0c2ea41be867f6ed612811963", size = 4423126, upload-time = "2025-10-15T23:18:04.85Z" },
     { url = "https://files.pythonhosted.org/packages/cd/c7/f65027c2810e14c3e7268353b1681932b87e5a48e65505d8cc17c99e36ae/cryptography-46.0.3-cp38-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:3b51b8ca4f1c6453d8829e1eb7299499ca7f313900dd4d89a24b8b87c0a780d4", size = 4686573, upload-time = "2025-10-15T23:18:06.908Z" },
@@ -672,7 +654,7 @@ name = "exceptiongroup"
 version = "1.3.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "typing-extensions", marker = "python_full_version < '3.13'" },
+    { name = "typing-extensions", marker = "python_full_version < '3.11'" },
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/0b/9f/a65090624ecf468cdca03533906e7c69ed7588582240cfe7cc9e770b50eb/exceptiongroup-1.3.0.tar.gz", hash = "sha256:b241f5885f560bc56a59ee63ca4c6a8bfa46ae4ad651af316d4e81817bb9fd88", size = 29749, upload-time = "2025-05-10T17:42:51.123Z" }
 wheels = [
@@ -863,7 +845,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/7d/ed/6bfa4109fcb23a58819600392564fea69cdc6551ffd5e69ccf1d52a40cbc/greenlet-3.2.4-cp310-cp310-macosx_11_0_universal2.whl", hash = "sha256:8c68325b0d0acf8d91dde4e6f930967dd52a5302cd4062932a6b2e7c2969f47c", size = 271061, upload-time = "2025-08-07T13:17:15.373Z" },
     { url = "https://files.pythonhosted.org/packages/2a/fc/102ec1a2fc015b3a7652abab7acf3541d58c04d3d17a8d3d6a44adae1eb1/greenlet-3.2.4-cp310-cp310-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:94385f101946790ae13da500603491f04a76b6e4c059dab271b3ce2e283b2590", size = 629475, upload-time = "2025-08-07T13:42:54.009Z" },
     { url = "https://files.pythonhosted.org/packages/c5/26/80383131d55a4ac0fb08d71660fd77e7660b9db6bdb4e8884f46d9f2cc04/greenlet-3.2.4-cp310-cp310-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:f10fd42b5ee276335863712fa3da6608e93f70629c631bf77145021600abc23c", size = 640802, upload-time = "2025-08-07T13:45:25.52Z" },
-    { url = "https://files.pythonhosted.org/packages/9f/7c/e7833dbcd8f376f3326bd728c845d31dcde4c84268d3921afcae77d90d08/greenlet-3.2.4-cp310-cp310-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:c8c9e331e58180d0d83c5b7999255721b725913ff6bc6cf39fa2a45841a4fd4b", size = 636703, upload-time = "2025-08-07T13:53:12.622Z" },
     { url = "https://files.pythonhosted.org/packages/e9/49/547b93b7c0428ede7b3f309bc965986874759f7d89e4e04aeddbc9699acb/greenlet-3.2.4-cp310-cp310-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:58b97143c9cc7b86fc458f215bd0932f1757ce649e05b640fea2e79b54cedb31", size = 635417, upload-time = "2025-08-07T13:18:25.189Z" },
     { url = "https://files.pythonhosted.org/packages/7f/91/ae2eb6b7979e2f9b035a9f612cf70f1bf54aad4e1d125129bef1eae96f19/greenlet-3.2.4-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:c2ca18a03a8cfb5b25bc1cbe20f3d9a4c80d8c3b13ba3df49ac3961af0b1018d", size = 584358, upload-time = "2025-08-07T13:18:23.708Z" },
     { url = "https://files.pythonhosted.org/packages/f7/85/433de0c9c0252b22b16d413c9407e6cb3b41df7389afc366ca204dbc1393/greenlet-3.2.4-cp310-cp310-musllinux_1_1_aarch64.whl", hash = "sha256:9fe0a28a7b952a21e2c062cd5756d34354117796c6d9215a87f55e38d15402c5", size = 1113550, upload-time = "2025-08-07T13:42:37.467Z" },
@@ -874,7 +855,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/a4/de/f28ced0a67749cac23fecb02b694f6473f47686dff6afaa211d186e2ef9c/greenlet-3.2.4-cp311-cp311-macosx_11_0_universal2.whl", hash = "sha256:96378df1de302bc38e99c3a9aa311967b7dc80ced1dcc6f171e99842987882a2", size = 272305, upload-time = "2025-08-07T13:15:41.288Z" },
     { url = "https://files.pythonhosted.org/packages/09/16/2c3792cba130000bf2a31c5272999113f4764fd9d874fb257ff588ac779a/greenlet-3.2.4-cp311-cp311-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:1ee8fae0519a337f2329cb78bd7a8e128ec0f881073d43f023c7b8d4831d5246", size = 632472, upload-time = "2025-08-07T13:42:55.044Z" },
     { url = "https://files.pythonhosted.org/packages/ae/8f/95d48d7e3d433e6dae5b1682e4292242a53f22df82e6d3dda81b1701a960/greenlet-3.2.4-cp311-cp311-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:94abf90142c2a18151632371140b3dba4dee031633fe614cb592dbb6c9e17bc3", size = 644646, upload-time = "2025-08-07T13:45:26.523Z" },
-    { url = "https://files.pythonhosted.org/packages/d5/5e/405965351aef8c76b8ef7ad370e5da58d57ef6068df197548b015464001a/greenlet-3.2.4-cp311-cp311-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:4d1378601b85e2e5171b99be8d2dc85f594c79967599328f95c1dc1a40f1c633", size = 640519, upload-time = "2025-08-07T13:53:13.928Z" },
     { url = "https://files.pythonhosted.org/packages/25/5d/382753b52006ce0218297ec1b628e048c4e64b155379331f25a7316eb749/greenlet-3.2.4-cp311-cp311-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:0db5594dce18db94f7d1650d7489909b57afde4c580806b8d9203b6e79cdc079", size = 639707, upload-time = "2025-08-07T13:18:27.146Z" },
     { url = "https://files.pythonhosted.org/packages/1f/8e/abdd3f14d735b2929290a018ecf133c901be4874b858dd1c604b9319f064/greenlet-3.2.4-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:2523e5246274f54fdadbce8494458a2ebdcdbc7b802318466ac5606d3cded1f8", size = 587684, upload-time = "2025-08-07T13:18:25.164Z" },
     { url = "https://files.pythonhosted.org/packages/5d/65/deb2a69c3e5996439b0176f6651e0052542bb6c8f8ec2e3fba97c9768805/greenlet-3.2.4-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:1987de92fec508535687fb807a5cea1560f6196285a4cde35c100b8cd632cc52", size = 1116647, upload-time = "2025-08-07T13:42:38.655Z" },
@@ -885,7 +865,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/44/69/9b804adb5fd0671f367781560eb5eb586c4d495277c93bde4307b9e28068/greenlet-3.2.4-cp312-cp312-macosx_11_0_universal2.whl", hash = "sha256:3b67ca49f54cede0186854a008109d6ee71f66bd57bb36abd6d0a0267b540cdd", size = 274079, upload-time = "2025-08-07T13:15:45.033Z" },
     { url = "https://files.pythonhosted.org/packages/46/e9/d2a80c99f19a153eff70bc451ab78615583b8dac0754cfb942223d2c1a0d/greenlet-3.2.4-cp312-cp312-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:ddf9164e7a5b08e9d22511526865780a576f19ddd00d62f8a665949327fde8bb", size = 640997, upload-time = "2025-08-07T13:42:56.234Z" },
     { url = "https://files.pythonhosted.org/packages/3b/16/035dcfcc48715ccd345f3a93183267167cdd162ad123cd93067d86f27ce4/greenlet-3.2.4-cp312-cp312-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:f28588772bb5fb869a8eb331374ec06f24a83a9c25bfa1f38b6993afe9c1e968", size = 655185, upload-time = "2025-08-07T13:45:27.624Z" },
-    { url = "https://files.pythonhosted.org/packages/31/da/0386695eef69ffae1ad726881571dfe28b41970173947e7c558d9998de0f/greenlet-3.2.4-cp312-cp312-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:5c9320971821a7cb77cfab8d956fa8e39cd07ca44b6070db358ceb7f8797c8c9", size = 649926, upload-time = "2025-08-07T13:53:15.251Z" },
     { url = "https://files.pythonhosted.org/packages/68/88/69bf19fd4dc19981928ceacbc5fd4bb6bc2215d53199e367832e98d1d8fe/greenlet-3.2.4-cp312-cp312-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:c60a6d84229b271d44b70fb6e5fa23781abb5d742af7b808ae3f6efd7c9c60f6", size = 651839, upload-time = "2025-08-07T13:18:30.281Z" },
     { url = "https://files.pythonhosted.org/packages/19/0d/6660d55f7373b2ff8152401a83e02084956da23ae58cddbfb0b330978fe9/greenlet-3.2.4-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:3b3812d8d0c9579967815af437d96623f45c0f2ae5f04e366de62a12d83a8fb0", size = 607586, upload-time = "2025-08-07T13:18:28.544Z" },
     { url = "https://files.pythonhosted.org/packages/8e/1a/c953fdedd22d81ee4629afbb38d2f9d71e37d23caace44775a3a969147d4/greenlet-3.2.4-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:abbf57b5a870d30c4675928c37278493044d7c14378350b3aa5d484fa65575f0", size = 1123281, upload-time = "2025-08-07T13:42:39.858Z" },
@@ -896,7 +875,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/49/e8/58c7f85958bda41dafea50497cbd59738c5c43dbbea5ee83d651234398f4/greenlet-3.2.4-cp313-cp313-macosx_11_0_universal2.whl", hash = "sha256:1a921e542453fe531144e91e1feedf12e07351b1cf6c9e8a3325ea600a715a31", size = 272814, upload-time = "2025-08-07T13:15:50.011Z" },
     { url = "https://files.pythonhosted.org/packages/62/dd/b9f59862e9e257a16e4e610480cfffd29e3fae018a68c2332090b53aac3d/greenlet-3.2.4-cp313-cp313-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:cd3c8e693bff0fff6ba55f140bf390fa92c994083f838fece0f63be121334945", size = 641073, upload-time = "2025-08-07T13:42:57.23Z" },
     { url = "https://files.pythonhosted.org/packages/f7/0b/bc13f787394920b23073ca3b6c4a7a21396301ed75a655bcb47196b50e6e/greenlet-3.2.4-cp313-cp313-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:710638eb93b1fa52823aa91bf75326f9ecdfd5e0466f00789246a5280f4ba0fc", size = 655191, upload-time = "2025-08-07T13:45:29.752Z" },
-    { url = "https://files.pythonhosted.org/packages/f2/d6/6adde57d1345a8d0f14d31e4ab9c23cfe8e2cd39c3baf7674b4b0338d266/greenlet-3.2.4-cp313-cp313-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:c5111ccdc9c88f423426df3fd1811bfc40ed66264d35aa373420a34377efc98a", size = 649516, upload-time = "2025-08-07T13:53:16.314Z" },
     { url = "https://files.pythonhosted.org/packages/7f/3b/3a3328a788d4a473889a2d403199932be55b1b0060f4ddd96ee7cdfcad10/greenlet-3.2.4-cp313-cp313-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:d76383238584e9711e20ebe14db6c88ddcedc1829a9ad31a584389463b5aa504", size = 652169, upload-time = "2025-08-07T13:18:32.861Z" },
     { url = "https://files.pythonhosted.org/packages/ee/43/3cecdc0349359e1a527cbf2e3e28e5f8f06d3343aaf82ca13437a9aa290f/greenlet-3.2.4-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:23768528f2911bcd7e475210822ffb5254ed10d71f4028387e5a99b4c6699671", size = 610497, upload-time = "2025-08-07T13:18:31.636Z" },
     { url = "https://files.pythonhosted.org/packages/b8/19/06b6cf5d604e2c382a6f31cafafd6f33d5dea706f4db7bdab184bad2b21d/greenlet-3.2.4-cp313-cp313-musllinux_1_1_aarch64.whl", hash = "sha256:00fadb3fedccc447f517ee0d3fd8fe49eae949e1cd0f6a611818f4f6fb7dc83b", size = 1121662, upload-time = "2025-08-07T13:42:41.117Z" },
@@ -907,7 +885,6 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/22/5c/85273fd7cc388285632b0498dbbab97596e04b154933dfe0f3e68156c68c/greenlet-3.2.4-cp314-cp314-macosx_11_0_universal2.whl", hash = "sha256:49a30d5fda2507ae77be16479bdb62a660fa51b1eb4928b524975b3bde77b3c0", size = 273586, upload-time = "2025-08-07T13:16:08.004Z" },
     { url = "https://files.pythonhosted.org/packages/d1/75/10aeeaa3da9332c2e761e4c50d4c3556c21113ee3f0afa2cf5769946f7a3/greenlet-3.2.4-cp314-cp314-manylinux2014_aarch64.manylinux_2_17_aarch64.whl", hash = "sha256:299fd615cd8fc86267b47597123e3f43ad79c9d8a22bebdce535e53550763e2f", size = 686346, upload-time = "2025-08-07T13:42:59.944Z" },
     { url = "https://files.pythonhosted.org/packages/c0/aa/687d6b12ffb505a4447567d1f3abea23bd20e73a5bed63871178e0831b7a/greenlet-3.2.4-cp314-cp314-manylinux2014_ppc64le.manylinux_2_17_ppc64le.whl", hash = "sha256:c17b6b34111ea72fc5a4e4beec9711d2226285f0386ea83477cbb97c30a3f3a5", size = 699218, upload-time = "2025-08-07T13:45:30.969Z" },
-    { url = "https://files.pythonhosted.org/packages/dc/8b/29aae55436521f1d6f8ff4e12fb676f3400de7fcf27fccd1d4d17fd8fecd/greenlet-3.2.4-cp314-cp314-manylinux2014_s390x.manylinux_2_17_s390x.whl", hash = "sha256:b4a1870c51720687af7fa3e7cda6d08d801dae660f75a76f3845b642b4da6ee1", size = 694659, upload-time = "2025-08-07T13:53:17.759Z" },
     { url = "https://files.pythonhosted.org/packages/92/2e/ea25914b1ebfde93b6fc4ff46d6864564fba59024e928bdc7de475affc25/greenlet-3.2.4-cp314-cp314-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:061dc4cf2c34852b052a8620d40f36324554bc192be474b9e9770e8c042fd735", size = 695355, upload-time = "2025-08-07T13:18:34.517Z" },
     { url = "https://files.pythonhosted.org/packages/72/60/fc56c62046ec17f6b0d3060564562c64c862948c9d4bc8aa807cf5bd74f4/greenlet-3.2.4-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:44358b9bf66c8576a9f57a590d5f5d6e72fa4228b763d0e43fee6d3b06d3a337", size = 657512, upload-time = "2025-08-07T13:18:33.969Z" },
     { url = "https://files.pythonhosted.org/packages/23/6e/74407aed965a4ab6ddd93a7ded3180b730d281c77b765788419484cdfeef/greenlet-3.2.4-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:2917bdf657f5859fbf3386b12d68ede4cf1f04c90c3a6bc1f013dd68a22e2269", size = 1612508, upload-time = "2025-11-04T12:42:23.427Z" },
@@ -2152,6 +2129,7 @@ wheels = [
 
 [[package]]
 name = "posthog"
+version = "7.9.7"
 source = { editable = "." }
 dependencies = [
     { name = "backoff" },