feat(ai): langchain cached and reasoning tokens (#258)

* fix: reasoning and cached tokens

* test: new flows

* fix: missing field

* chore: bump

* fix: make sure we send write/read/reasoning tokens

.github/workflows/call-flags-project-board.yml

@@ -0,0 +1,17 @@
+# This workflow is used to call the flags-project-board workflow when a pull request is opened, ready for review, review requested, synchronized, converted to draft, or reopened.
+# It is used to update the feature flags project board with the pull request information.
+
+name: Call Feature Flags Project Workflow
+
+on:
+    pull_request:
+        types: [opened, ready_for_review, review_requested, synchronize, converted_to_draft, reopened]
+
+jobs:
+    call-flags-project:
+        uses: PostHog/.github/.github/workflows/flags-project-board.yml@main
+        with:
+            pr_number: ${{ github.event.pull_request.number }}
+            pr_node_id: ${{ github.event.pull_request.node_id }}
+            is_draft: ${{ github.event.pull_request.draft }}
+        secrets: inherit

.github/workflows/ci.yml

@@ -18,17 +18,16 @@ jobs:
               with:
                   python-version: 3.11.11
 
-            - uses: actions/cache@5a3ec84eff668545956fd18022155c47e93e2684
+            - name: Install uv
+              uses: astral-sh/setup-uv@0c5e2b8115b80b4c7c5ddf6ffdd634974642d182 # v5.4.1
               with:
-                  path: ~/.cache/pip
-                  key: ${{ runner.os }}-pip-${{ hashFiles('pyproject.toml') }}
-                  restore-keys: |
-                      ${{ runner.os }}-pip-
+                  enable-cache: true
+                  pyproject-file: 'pyproject.toml'
 
             - name: Install dev dependencies
+              shell: bash
               run: |
-                  python -m pip install -e .[dev]
-              if: steps.cache.outputs.cache-hit != 'true'
+                  UV_PROJECT_ENVIRONMENT=$pythonLocation uv sync --extra dev
 
             - name: Check formatting with ruff
               run: |
@@ -55,9 +54,16 @@ jobs:
               with:
                   python-version: ${{ matrix.python-version }}
 
-            - name: Install requirements.txt dependencies with pip
+            - name: Install uv
+              uses: astral-sh/setup-uv@0c5e2b8115b80b4c7c5ddf6ffdd634974642d182 # v5.4.1
+              with:
+                  enable-cache: true
+                  pyproject-file: 'pyproject.toml'
+
+            - name: Install test dependencies
+              shell: bash
               run: |
-                  python -m pip install -e .[test]
+                  UV_PROJECT_ENVIRONMENT=$pythonLocation uv sync --extra test
 
             - name: Run posthog tests
               run: |

.github/workflows/release.yaml

@@ -24,15 +24,23 @@ jobs:
 
       - name: Set up Python
         uses: actions/setup-python@8d9ed9ac5c53483de85588cdf95a591a75ab9f55
+        with:
+          python-version: 3.11.11
 
+      - name: Install uv
+        uses: astral-sh/setup-uv@0c5e2b8115b80b4c7c5ddf6ffdd634974642d182 # v5.4.1
+        with:
+            enable-cache: true
+            pyproject-file: 'pyproject.toml'
+            
       - name: Detect version
         run: echo "REPO_VERSION=$(python3 posthog/version.py)" >> $GITHUB_ENV
 
       - name: Prepare for building release
-        run: pip install -U pip setuptools packaging wheel twine
+        run: uv sync --extra dev
 
-      - name: Push release to PyPI
-        run: make release && make release_analytics
+      - name: Push releases to PyPI
+        run: uv run make release && uv run make release_analytics
 
       - name: Create GitHub release
         uses: actions/create-release@0cb9c9b65d5d1901c1f53e5e66eaf4afd303e70e # v1

.gitignore

@@ -17,3 +17,4 @@ posthog-analytics
 .coverage
 pyrightconfig.json
 .env
+.DS_Store

BEFORE_SEND.md

@@ -0,0 +1,237 @@
+# Before Send Hook
+
+The `before_send` parameter allows you to modify or filter events before they are sent to PostHog. This is useful for:
+
+- **Privacy**: Removing or masking sensitive data (PII)
+- **Filtering**: Dropping unwanted events (test events, internal users, etc.)
+- **Enhancement**: Adding custom properties to all events
+- **Transformation**: Modifying event names or property formats
+
+## Basic Usage
+
+```python
+import posthog
+from typing import Optional, Dict, Any
+
+def my_before_send(event: Dict[str, Any]) -> Optional[Dict[str, Any]]:
+    """
+    Process event before sending to PostHog.
+    
+    Args:
+        event: The event dictionary containing 'event', 'distinct_id', 'properties', etc.
+    
+    Returns:
+        Modified event dictionary to send, or None to drop the event
+    """
+    # Your processing logic here
+    return event
+
+# Initialize client with before_send hook
+client = posthog.Client(
+    api_key="your-project-api-key",
+    before_send=my_before_send
+)
+```
+
+## Common Use Cases
+
+### 1. Filter Out Events
+
+```python
+from typing import Optional, Any
+
+def filter_events_by_property_or_event_name(event: dict[str, Any]) -> Optional[dict[str, Any]]:
+    """Drop events from internal users or test environments."""
+    properties = event.get("properties", {})
+    
+    # Choose some property from your events
+    event_source = properties.get("event_source", "")
+    if event_source.endswith("internal"):
+        return None  # Drop the event
+    
+    # Filter out test events
+    if event.get("event") == "test_event":
+        return None
+    
+    return event
+```
+
+### 2. Remove/Mask PII Data
+
+```python
+from typing import Optional, Any
+
+def scrub_pii(event: dict[str, Any]) -> Optional[dict[str, Any]]:
+    """Remove or mask personally identifiable information."""
+    properties = event.get("properties", {})
+    
+    # Mask email but keep domain for analytics
+    if "email" in properties:
+        email = properties["email"]
+        if "@" in email:
+            domain = email.split("@")[1]
+            properties["email"] = f"***@{domain}"
+        else:
+            properties["email"] = "***"
+    
+    # Remove sensitive fields entirely
+    sensitive_fields = ["my_business_info", "secret_things"]
+    for field in sensitive_fields:
+        properties.pop(field, None)
+    
+    return event
+```
+
+### 3. Add Custom Properties
+
+```python
+from typing import Optional, Any
+
+from datetime import datetime
+from typing import Optional, Any
+
+def add_context(event: dict[str, Any]) -> Optional[dict[str, Any]]:
+    """Add custom properties to all events."""
+    if "properties" not in event:
+        event["properties"] = {}
+    
+    event["properties"].update({
+        "app_version": "2.1.0",
+        "environment": "production", 
+        "processed_at": datetime.now().isoformat()
+    })
+    
+    return event
+```
+
+### 4. Transform Event Names
+
+```python
+from typing import Optional, Any
+
+def normalize_event_names(event: dict[str, Any]) -> Optional[dict[str, Any]]:
+    """Convert event names to a consistent format."""
+    original_event = event.get("event")
+    if original_event:
+        # Convert to snake_case
+        normalized = original_event.lower().replace(" ", "_").replace("-", "_")
+        event["event"] = f"app_{normalized}"
+    
+    return event
+```
+
+### 5. Log and drop in "dev" mode
+
+When running in local dev often, you want to log but drop all events
+
+
+```python
+from typing import Optional, Any
+
+def log_and_drop_all(event: dict[str, Any]) -> Optional[dict[str, Any]]:
+    """Convert event names to a consistent format."""
+    print(event)
+    
+    return None
+```
+
+### 6. Combined Processing
+
+```python
+from typing import Optional, Any
+
+def comprehensive_processor(event: dict[str, Any]) -> Optional[dict[str, Any]]:
+    """Apply multiple transformations in sequence."""
+    
+    # Step 1: Filter unwanted events
+    if should_drop_event(event):
+        return None
+    
+    # Step 2: Scrub PII
+    event = scrub_pii(event)
+    
+    # Step 3: Add context
+    event = add_context(event)
+    
+    # Step 4: Normalize names
+    event = normalize_event_names(event)
+    
+    return event
+
+def should_drop_event(event: dict[str, Any]) -> bool:
+    """Determine if event should be dropped."""
+    # Your filtering logic
+    return False
+```
+
+## Error Handling
+
+If your `before_send` function raises an exception, PostHog will:
+
+1. Log the error 
+2. Continue with the original, unmodified event
+3. Not crash your application
+
+```python
+from typing import Optional, Any
+
+def risky_before_send(event: dict[str, Any]) -> Optional[dict[str, Any]]:
+    # If this raises an exception, the original event will be sent
+    risky_operation()
+    return event
+```
+
+## Complete Example
+
+```python
+import posthog
+from typing import Optional, Any
+import re
+
+def production_before_send(event: dict[str, Any]) -> Optional[dict[str, Any]]:
+    try:
+        properties = event.get("properties", {})
+        
+        # 1. Filter out bot traffic
+        user_agent = properties.get("$user_agent", "")
+        if re.search(r'bot|crawler|spider', user_agent, re.I):
+            return None
+        
+        # 2. Filter out internal traffic  
+        ip = properties.get("$ip", "")
+        if ip.startswith("192.168.") or ip.startswith("10."):
+            return None
+        
+        # 3. Scrub email PII but keep domain
+        if "email" in properties:
+            email = properties["email"]
+            if "@" in email:
+                domain = email.split("@")[1]
+                properties["email"] = f"***@{domain}"
+        
+        # 4. Add custom context
+        properties.update({
+            "app_version": "1.0.0",
+            "build_number": "123"
+        })
+        
+        # 5. Normalize event name
+        if event.get("event"):
+            event["event"] = event["event"].lower().replace(" ", "_")
+        
+        return event
+        
+    except Exception as e:
+        # Log error but don't crash
+        print(f"Error in before_send: {e}")
+        return event  # Return original event on error
+
+# Usage
+client = posthog.Client(
+    api_key="your-api-key",
+    before_send=production_before_send
+)
+
+# All events will now be processed by your before_send function
+client.capture("user_123", "Page View", {"url": "/home"})
+```

CHANGELOG.md

@@ -1,10 +1,48 @@
-## 4.3.3 - 2025-06-06
+## 4.9.0 - 2025-06-13
 
-Add `setup()` function to initialise default client
+- feat(ai): track reasoning and cache tokens in the LangChain callback
+
+## 4.8.0 - 2025-06-10
+
+- fix: export scoped, rather than tracked, decorator
+- feat: allow use of contexts without error tracking
+
+## 4.7.0 - 2025-06-10
+
+- feat: add support for parse endpoint in responses API (no longer beta)
+
+## 4.6.2 - 2025-06-09
+
+- fix: replace `import posthog` with direct method imports
+
+## 4.6.1 - 2025-06-09
+
+- fix: replace `import posthog` in `posthoganalytics` package
+
+## 4.6.0 - 2025-06-09
+
+- feat: add additional user and request context to captured exceptions via the Django integration
+- feat: Add `setup()` function to initialise default client
+
+## 4.5.0 - 2025-06-09
+
+- feat: add before_send callback (#249)
+
+## 4.4.2- 2025-06-09
+
+- empty point release to fix release automation
+
+## 4.4.1 2025-06-09
+
+- empty point release to fix release automation
+
+## 4.4.0 - 2025-06-09
+
+- Use the new `/flags` endpoint for all feature flag evaluations (don't fall back to `/decide` at all)
 
 ## 4.3.2 - 2025-06-06
 
-Add context management:
+1. Add context management:
 
 - New context manager with `posthog.new_context()`
 - Tag functions: `posthog.tag()`, `posthog.get_tags()`, `posthog.clear_tags()`
@@ -12,12 +50,10 @@ Add context management:
   - `@posthog.scoped` - Creates context and captures exceptions thrown within the function
 - Automatic deduplication of exceptions to ensure each exception is only captured once
 
-## 4.2.1 - 2025-6-05
-
-1. fix: feature flag request use geoip_disable (#235)
-2. chore: pin actions versions (#210)
-3. fix: opinionated setup and clean fn fix (#240)
-4. fix: release action failed (#241)
+2. fix: feature flag request use geoip_disable (#235)
+3. chore: pin actions versions (#210)
+4. fix: opinionated setup and clean fn fix (#240)
+5. fix: release action failed (#241)
 
 ## 4.2.0 - 2025-05-22
 

Makefile

@@ -16,16 +16,21 @@ release_analytics:
 	rm -rf posthoganalytics
 	mkdir posthoganalytics
 	cp -r posthog/* posthoganalytics/
-	find ./posthoganalytics -type f -exec sed -i '' -e 's/from posthog /from posthoganalytics /g' {} \;
-	find ./posthoganalytics -type f -exec sed -i '' -e 's/from posthog\./from posthoganalytics\./g' {} \;
+	find ./posthoganalytics -type f -name "*.py" -exec sed -i.bak -e 's/from posthog /from posthoganalytics /g' {} \;
+	find ./posthoganalytics -type f -name "*.py" -exec sed -i.bak -e 's/from posthog\./from posthoganalytics\./g' {} \;
+	find ./posthoganalytics -name "*.bak" -delete
 	rm -rf posthog
 	python setup_analytics.py sdist bdist_wheel
 	twine upload dist/*
 	mkdir posthog
-	find ./posthoganalytics -type f -exec sed -i '' -e 's/from posthoganalytics /from posthog /g' {} \;
-	find ./posthoganalytics -type f -exec sed -i '' -e 's/from posthoganalytics\./from posthog\./g' {} \;
+	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' {} \;
+	find ./posthoganalytics -name "*.bak" -delete
 	cp -r posthoganalytics/* posthog/
 	rm -rf posthoganalytics
+	rm -f pyproject.toml
+	cp pyproject.toml.backup pyproject.toml
+	rm -f pyproject.toml.backup
 
 e2e_test:
 	.buildscripts/e2e.sh

README.md

@@ -16,11 +16,13 @@ Please see the [Python integration docs](https://posthog.com/docs/integrations/p
 
 ### Testing Locally
 
-1. Run `python3 -m venv env` (creates virtual environment called "env")
-    * or `uv venv env`
+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`
 2. Run `source env/bin/activate` (activates the virtual environment)
-3. Run `python3 -m pip install -e ".[test]"` (installs the package in develop mode, along with test dependencies)
-    * or `uv pip install -e ".[test]"`
+3. Run `uv sync --extra dev --extra test` (installs the package in develop mode, along with test dependencies)
+    * 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`
@@ -32,7 +34,7 @@ uv python install 3.9.19
 uv python pin 3.9.19
 uv venv env
 source env/bin/activate
-uv pip install --editable ".[dev,test]"
+uv sync --extra dev --extra test
 pre-commit install
 make test
 ```

posthog/__init__.py

@@ -15,7 +15,7 @@ new_context = new_context
 tag = tag
 get_tags = get_tags
 clear_tags = clear_tags
-tracked = scoped
+scoped = scoped
 
 """Settings."""
 api_key = None  # type: Optional[str]

posthog/ai/langchain/callbacks.py

@@ -14,7 +14,6 @@ from typing import (
     List,
     Optional,
     Sequence,
-    Tuple,
     Union,
     cast,
 )
@@ -569,9 +568,14 @@ class CallbackHandler(BaseCallbackHandler):
             event_properties["$ai_is_error"] = True
         else:
             # Add usage
-            input_tokens, output_tokens = _parse_usage(output)
-            event_properties["$ai_input_tokens"] = input_tokens
-            event_properties["$ai_output_tokens"] = output_tokens
+            usage = _parse_usage(output)
+            event_properties["$ai_input_tokens"] = usage.input_tokens
+            event_properties["$ai_output_tokens"] = usage.output_tokens
+            event_properties["$ai_cache_creation_input_tokens"] = (
+                usage.cache_write_tokens
+            )
+            event_properties["$ai_cache_read_input_tokens"] = usage.cache_read_tokens
+            event_properties["$ai_reasoning_tokens"] = usage.reasoning_tokens
 
             # Generation results
             generation_result = output.generations[-1]
@@ -647,9 +651,18 @@ def _convert_message_to_dict(message: BaseMessage) -> Dict[str, Any]:
     return message_dict
 
 
+@dataclass
+class ModelUsage:
+    input_tokens: Optional[int]
+    output_tokens: Optional[int]
+    cache_write_tokens: Optional[int]
+    cache_read_tokens: Optional[int]
+    reasoning_tokens: Optional[int]
+
+
 def _parse_usage_model(
-    usage: Union[BaseModel, Dict],
-) -> Tuple[Union[int, None], Union[int, None]]:
+    usage: Union[BaseModel, dict],
+) -> ModelUsage:
     if isinstance(usage, BaseModel):
         usage = usage.__dict__
 
@@ -657,15 +670,23 @@ def _parse_usage_model(
         # https://pypi.org/project/langchain-anthropic/ (works also for Bedrock-Anthropic)
         ("input_tokens", "input"),
         ("output_tokens", "output"),
+        ("cache_creation_input_tokens", "cache_write"),
+        ("cache_read_input_tokens", "cache_read"),
         # https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/get-token-count
         ("prompt_token_count", "input"),
         ("candidates_token_count", "output"),
+        ("cached_content_token_count", "cache_read"),
+        ("thoughts_token_count", "reasoning"),
         # Bedrock: https://docs.aws.amazon.com/bedrock/latest/userguide/monitoring-cw.html#runtime-cloudwatch-metrics
         ("inputTokenCount", "input"),
         ("outputTokenCount", "output"),
+        ("cacheCreationInputTokenCount", "cache_write"),
+        ("cacheReadInputTokenCount", "cache_read"),
         # Bedrock Anthropic
         ("prompt_tokens", "input"),
         ("completion_tokens", "output"),
+        ("cache_creation_input_tokens", "cache_write"),
+        ("cache_read_input_tokens", "cache_read"),
         # langchain-ibm https://pypi.org/project/langchain-ibm/
         ("input_token_count", "input"),
         ("generated_token_count", "output"),
@@ -683,13 +704,45 @@ def _parse_usage_model(
 
             parsed_usage[type_key] = final_count
 
-    return parsed_usage.get("input"), parsed_usage.get("output")
+    # Caching (OpenAI & langchain 0.3.9+)
+    if "input_token_details" in usage and isinstance(
+        usage["input_token_details"], dict
+    ):
+        parsed_usage["cache_write"] = usage["input_token_details"].get("cache_creation")
+        parsed_usage["cache_read"] = usage["input_token_details"].get("cache_read")
+
+    # Reasoning (OpenAI & langchain 0.3.9+)
+    if "output_token_details" in usage and isinstance(
+        usage["output_token_details"], dict
+    ):
+        parsed_usage["reasoning"] = usage["output_token_details"].get("reasoning")
+
+    field_mapping = {
+        "input": "input_tokens",
+        "output": "output_tokens",
+        "cache_write": "cache_write_tokens",
+        "cache_read": "cache_read_tokens",
+        "reasoning": "reasoning_tokens",
+    }
+    return ModelUsage(
+        **{
+            dataclass_key: parsed_usage.get(mapped_key) or 0
+            for mapped_key, dataclass_key in field_mapping.items()
+        },
+    )
 
 
-def _parse_usage(response: LLMResult):
+def _parse_usage(response: LLMResult) -> ModelUsage:
     # langchain-anthropic uses the usage field
     llm_usage_keys = ["token_usage", "usage"]
-    llm_usage: Tuple[Union[int, None], Union[int, None]] = (None, None)
+    llm_usage: ModelUsage = ModelUsage(
+        input_tokens=None,
+        output_tokens=None,
+        cache_write_tokens=None,
+        cache_read_tokens=None,
+        reasoning_tokens=None,
+    )
+
     if response.llm_output is not None:
         for key in llm_usage_keys:
             if response.llm_output.get(key):

posthog/ai/openai/openai.py

@@ -230,6 +230,42 @@ class WrappedResponses:
                 groups=posthog_groups,
             )
 
+    def parse(
+        self,
+        posthog_distinct_id: Optional[str] = None,
+        posthog_trace_id: Optional[str] = None,
+        posthog_properties: Optional[Dict[str, Any]] = None,
+        posthog_privacy_mode: bool = False,
+        posthog_groups: Optional[Dict[str, Any]] = None,
+        **kwargs: Any,
+    ):
+        """
+        Parse structured output using OpenAI's 'responses.parse' method, but also track usage in PostHog.
+
+        Args:
+            posthog_distinct_id: Optional ID to associate with the usage event.
+            posthog_trace_id: Optional trace UUID for linking events.
+            posthog_properties: Optional dictionary of extra properties to include in the event.
+            posthog_privacy_mode: Whether to anonymize the input and output.
+            posthog_groups: Optional dictionary of groups to associate with the event.
+            **kwargs: Any additional parameters for the OpenAI Responses Parse API.
+
+        Returns:
+            The response from OpenAI's responses.parse call.
+        """
+        return call_llm_and_track_usage(
+            posthog_distinct_id,
+            self._client._ph_client,
+            "openai",
+            posthog_trace_id,
+            posthog_properties,
+            posthog_privacy_mode,
+            posthog_groups,
+            self._client.base_url,
+            self._original.parse,
+            **kwargs,
+        )
+
 
 class WrappedChat:
     """Wrapper for OpenAI chat that tracks usage in PostHog."""

posthog/ai/openai/openai_async.py

@@ -230,6 +230,42 @@ class WrappedResponses:
                 groups=posthog_groups,
             )
 
+    async def parse(
+        self,
+        posthog_distinct_id: Optional[str] = None,
+        posthog_trace_id: Optional[str] = None,
+        posthog_properties: Optional[Dict[str, Any]] = None,
+        posthog_privacy_mode: bool = False,
+        posthog_groups: Optional[Dict[str, Any]] = None,
+        **kwargs: Any,
+    ):
+        """
+        Parse structured output using OpenAI's 'responses.parse' method, but also track usage in PostHog.
+
+        Args:
+            posthog_distinct_id: Optional ID to associate with the usage event.
+            posthog_trace_id: Optional trace UUID for linking events.
+            posthog_properties: Optional dictionary of extra properties to include in the event.
+            posthog_privacy_mode: Whether to anonymize the input and output.
+            posthog_groups: Optional dictionary of groups to associate with the event.
+            **kwargs: Any additional parameters for the OpenAI Responses Parse API.
+
+        Returns:
+            The response from OpenAI's responses.parse call.
+        """
+        return await call_llm_and_track_usage_async(
+            posthog_distinct_id,
+            self._client._ph_client,
+            "openai",
+            posthog_trace_id,
+            posthog_properties,
+            posthog_privacy_mode,
+            posthog_groups,
+            self._client.base_url,
+            self._original.parse,
+            **kwargs,
+        )
+
 
 class WrappedChat:
     """Async wrapper for OpenAI chat that tracks usage in PostHog."""

posthog/client.py

@@ -1,5 +1,4 @@
 import atexit
-import hashlib
 import logging
 import numbers
 import os
@@ -27,7 +26,6 @@ from posthog.request import (
     DEFAULT_HOST,
     APIError,
     batch_post,
-    decide,
     determine_server_host,
     flags,
     get,
@@ -58,78 +56,6 @@ except ImportError:
 ID_TYPES = (numbers.Number, string_types, UUID)
 MAX_DICT_SIZE = 50_000
 
-# TODO: Get rid of these when you're done rolling out `/flags` to all customers
-ROLLOUT_PERCENTAGE = 1
-INCLUDED_HASHES = set(
-    {"bc94e67150c97dbcbf52549d50a7b80814841dbf"}
-)  # this is PostHog's API key
-# Explicitly excluding all the API tokens associated with the top 10 customers; we'll get to them soon, but don't want to rollout to them just yet
-EXCLUDED_HASHES = set(
-    {
-        "03005596796f9ee626e9596b8062972cb6a556a0",
-        "05620a20b287e0d5cb1d4a0dd492797f36b952c5",
-        "0f95b5ca12878693c01c6420e727904f1737caa7",
-        "1212b6287a6e7e5ff6be5cb30ec563f35c2139d6",
-        "171ec1bb2caf762e06b1fde2e36a38c4638691a8",
-        "171faa9fc754b1aa42252a4eedb948b7c805d5cb",
-        "178ddde3f628fb0030321387acf939e4e6946d35",
-        "1790085d7e9aa136e8b73c180dd6a6060e2ef949",
-        "1895a3349c2371559c886f19ef1bf60617a934e0",
-        "1f01267d4f0295f88e8943bc963d816ee4abc84b",
-        "213df54990a34e62e3570b430f7ee36ec0928743",
-        "23d235537d988ab98ad259853eab02b07d828c2b",
-        "27135f7ae8f936222a5fcfcdc75c139b27dd3254",
-        "2817396d80fafc86c0816af8e73880f8b3e54320",
-        "29d3235e63db42056858ef04c6a5488c2a459eaa",
-        "2a76d9b5eb9307e540de9d516aa80f6cb5a0292f",
-        "2a92965a1344ab8a1f7dac2507e858f579a88ac2",
-        "2d5823818261512d616161de2bb8a161d48f1e35",
-        "32942f6a879dbfa8011cc68288c098e4a76e6cc0",
-        "3db6c17ab65827ceadf77d9a8462fabd94170ca6",
-        "4975b24f9ced9b2c06b604ddc9612f663f9452d5",
-        "497c7b017b13cd6cdbfe641c71f0dfb660a4c518",
-        "49c79e1dbce4a7b9394d6c14bf0421e04cecb445",
-        "4d63e1c5cd3a80972eac4e7526f03357ac538043",
-        "4da0f42a6f8f116822411152e5cda3c65ed2561f",
-        "4e494675ecd2b841784d6f29b658b38a0877a62e",
-        "4e852d8422130cec991eca2d6416dbe321d0a689",
-        "5120bfd92c9c6731074a89e4a82f49e947d34369",
-        "512cd72f9aa7ab11dfd012cc2e19394a020bd9a8",
-        "5b175d4064cc62f01118a2c6818c2c02fc8f27e1",
-        "5ba4bba3979e97d2c84df2aba394ca29c6c43187",
-        "639014946463614353ca640b268dc6592f62b652",
-        "643b9be9d50104e2b4ba94bc56688adba69c80fe",
-        "658f92992af9fc6a360143d72d93a36f63bbccb0",
-        "673a59c99739dfcee35202e428dd020b94866d52",
-        "67a9829b4997f5c6f3ab8173ad299f634adcfa53",
-        "6d686043e914ae8275df65e1ad890bd32a3b6fdd",
-        "6e4b5e1d649ad006d78f1f1617a9a0f35fc73078",
-        "6f1fc3a8fa9df54d00cbc1ef9ad5f24640589fd0",
-        "764e5fec2c7899cfee620fae8450fcc62cd72bf0",
-        "80ea6d6ed9a5895633c7bee7aba4323eeacdc90e",
-        "872e420156f583bc97351f3d83c02dae734a85df",
-        "8a24844cbeae31e74b4372964cdea74e99d9c0e2",
-        "975ae7330506d4583b000f96ad87abb41a0141ce",
-        "9e3d71378b340def3080e0a3a785a1b964cf43ef",
-        "9ede7b21365661331d024d92915de6e69749892b",
-        "a1ed1b4216ef4cec542c6b3b676507770be24ddc",
-        "a4f66a70a9647b3b89fc59f7642af8ffab073ba1",
-        "a7adb80be9e90948ab6bb726cc6e8e52694aec74",
-        "bca4b14ac8de49cccc02306c7bb6e5ae2acc0f72",
-        "bde5fe49f61e13629c5498d7428a7f6215e482a6",
-        "c54a7074c323aa7c5cb7b24bf826751b2a58f5d8",
-        "c552d20da0c87fb4ebe2da97c7f95c05eef2bca1",
-        "d7682f2d268f3064d433309af34f2935810989d2",
-        "d794ac43d8be26bf99f369ea79501eb774fe1b16",
-        "e0963e2552af77d46bb24d5b5806b5b456c64c5f",
-        "e6f14b2100cb0598925958b097ace82486037a25",
-        "e79ec399ad45f44a4295a5bb1322e2f14600ae39",
-        "eecf29f73f9c31009e5737a6c5ec3f87ec5b8ea6",
-        "f2c01f3cc770c7788257ee60910e2530f92eefc3",
-        "f7bbc58f4122b1e2812c0f1962c584cb404a1ac3",
-    }
-)
-
 
 def get_os_info():
     """
@@ -185,43 +111,6 @@ def system_context() -> dict[str, Any]:
     }
 
 
-def is_token_in_rollout(
-    token: str,
-    percentage: float = 0,
-    included_hashes: Optional[set[str]] = None,
-    excluded_hashes: Optional[set[str]] = None,
-) -> bool:
-    """
-    Determines if a token should be included in a rollout based on:
-    1. If its hash matches any included_hashes provided
-    2. If its hash falls within the percentage rollout
-
-    Args:
-        token: String to hash (usually API key)
-        percentage: Float between 0 and 1 representing rollout percentage
-        included_hashes: Optional set of specific SHA1 hashes to match against
-        excluded_hashes: Optional set of specific SHA1 hashes to exclude from rollout
-    Returns:
-        bool: True if token should be included in rollout
-    """
-    # First generate SHA1 hash of token
-    token_hash = hashlib.sha1(token.encode("utf-8")).hexdigest()
-
-    # Check if hash matches any included hashes
-    if included_hashes and token_hash in included_hashes:
-        return True
-
-    # Check if hash matches any excluded hashes
-    if excluded_hashes and token_hash in excluded_hashes:
-        return False
-
-    # Convert first 8 chars of hash to int and divide by max value to get number between 0-1
-    hash_int = int(token_hash[:8], 16)
-    hash_float = hash_int / 0xFFFFFFFF
-
-    return hash_float < percentage
-
-
 class Client(object):
     """Create a new PostHog client."""
 
@@ -255,6 +144,7 @@ class Client(object):
         exception_autocapture_integrations=None,
         project_root=None,
         privacy_mode=False,
+        before_send=None,
     ):
         self.queue = queue.Queue(max_queue_size)
 
@@ -310,6 +200,15 @@ class Client(object):
         else:
             self.log.setLevel(logging.WARNING)
 
+        if before_send is not None:
+            if callable(before_send):
+                self.before_send = before_send
+            else:
+                self.log.warning("before_send is not callable, it will be ignored")
+                self.before_send = None
+        else:
+            self.before_send = None
+
         if self.enable_exception_autocapture:
             self.exception_capture = ExceptionCapture(
                 self, integrations=self.exception_autocapture_integrations
@@ -475,28 +374,13 @@ class Client(object):
             "geoip_disable": disable_geoip,
         }
 
-        use_flags = is_token_in_rollout(
+        resp_data = flags(
             self.api_key,
-            ROLLOUT_PERCENTAGE,
-            included_hashes=INCLUDED_HASHES,
-            excluded_hashes=EXCLUDED_HASHES,
+            self.host,
+            timeout=self.feature_flags_request_timeout_seconds,
+            **request_data,
         )
 
-        if use_flags:
-            resp_data = flags(
-                self.api_key,
-                self.host,
-                timeout=self.feature_flags_request_timeout_seconds,
-                **request_data,
-            )
-        else:
-            resp_data = decide(
-                self.api_key,
-                self.host,
-                timeout=self.feature_flags_request_timeout_seconds,
-                **request_data,
-            )
-
         return normalize_flags_response(resp_data)
 
     def capture(
@@ -870,6 +754,18 @@ class Client(object):
         msg["distinct_id"] = stringify_id(msg.get("distinct_id", None))
 
         msg = clean(msg)
+
+        if self.before_send:
+            try:
+                modified_msg = self.before_send(msg)
+                if modified_msg is None:
+                    self.log.debug("Event dropped by before_send callback")
+                    return True, None
+                msg = modified_msg
+            except Exception as e:
+                self.log.exception(f"Error in before_send callback: {e}")
+                # Continue with the original message if callback fails
+
         self.log.debug("queueing: %s", msg)
 
         # if send is False, return msg as if it was successfully queued

posthog/exception_integrations/django.py

@@ -67,8 +67,8 @@ class DjangoRequestExtractor:
         headers = self.headers()
 
         # Extract traceparent and tracestate headers
-        traceparent = headers.get("traceparent")
-        tracestate = headers.get("tracestate")
+        traceparent = headers.get("Traceparent")
+        tracestate = headers.get("Tracestate")
 
         # Extract the distinct_id from tracestate
         distinct_id = None
@@ -80,12 +80,38 @@ class DjangoRequestExtractor:
                 distinct_id = match.group(1)
 
         return {
+            **self.user(),
             "distinct_id": distinct_id,
             "ip": headers.get("X-Forwarded-For"),
             "user_agent": headers.get("User-Agent"),
             "traceparent": traceparent,
+            "$request_path": self.request.path,
         }
 
+    def user(self):
+        user_data: dict[str, str] = {}
+
+        user = getattr(self.request, "user", None)
+
+        if user is None or not user.is_authenticated:
+            return user_data
+
+        try:
+            user_id = str(user.pk)
+            if user_id:
+                user_data.setdefault("$user_id", user_id)
+        except Exception:
+            pass
+
+        try:
+            email = str(user.email)
+            if email:
+                user_data.setdefault("email", email)
+        except Exception:
+            pass
+
+        return user_data
+
     def headers(self):
         # type: () -> Dict[str, str]
         return dict(self.request.headers)

posthog/scopes.py

@@ -12,34 +12,37 @@ def _get_current_context() -> Dict[str, Any]:
 
 
 @contextmanager
-def new_context(fresh=False):
+def new_context(fresh=False, capture_exceptions=True):
     """
-     Create a new context scope that will be active for the duration of the with block.
-     Any tags set within this scope will be isolated to this context. Any exceptions raised
+    Create a new context scope that will be active for the duration of the with block.
+    Any tags set within this scope will be isolated to this context. Any exceptions raised
     or events captured within the context will be tagged with the context tags.
 
-     Args:
-         fresh: Whether to start with a fresh context (default: False).
-                If False, inherits tags from parent context.
-                If True, starts with no tags.
+    Args:
+        fresh: Whether to start with a fresh context (default: False).
+               If False, inherits tags from parent context.
+               If True, starts with no tags.
+        capture_exceptions: Whether to capture exceptions raised within the context (default: True).
+               If True, captures exceptions and tags them with the context tags before propagating them.
+               If False, exceptions will propagate without being tagged or captured.
 
-     Examples:
-         # Inherit parent context tags
-         with posthog.new_context():
-             posthog.tag("request_id", "123")
-             # Both this event and the exception will be tagged with the context tags
-             posthog.capture("event_name", {"property": "value"})
-             raise ValueError("Something went wrong")
+    Examples:
+        # Inherit parent context tags
+        with posthog.new_context():
+            posthog.tag("request_id", "123")
+            # Both this event and the exception will be tagged with the context tags
+            posthog.capture("event_name", {"property": "value"})
+            raise ValueError("Something went wrong")
 
-         # Start with fresh context (no inherited tags)
-         with posthog.new_context(fresh=True):
-             posthog.tag("request_id", "123")
-             # Both this event and the exception will be tagged with the context tags
-             posthog.capture("event_name", {"property": "value"})
-             raise ValueError("Something went wrong")
+        # Start with fresh context (no inherited tags)
+        with posthog.new_context(fresh=True):
+            posthog.tag("request_id", "123")
+            # Both this event and the exception will be tagged with the context tags
+            posthog.capture("event_name", {"property": "value"})
+            raise ValueError("Something went wrong")
 
     """
-    import posthog
+    from posthog import capture_exception
 
     current_tags = _get_current_context().copy()
     current_stack = _context_stack.get()
@@ -49,7 +52,8 @@ def new_context(fresh=False):
     try:
         yield
     except Exception as e:
-        posthog.capture_exception(e)
+        if capture_exceptions:
+            capture_exception(e)
         raise
     finally:
         _context_stack.reset(token)
@@ -88,13 +92,14 @@ def clear_tags() -> None:
 F = TypeVar("F", bound=Callable[..., Any])
 
 
-def scoped(fresh=False):
+def scoped(fresh=False, capture_exceptions=True):
     """
     Decorator that creates a new context for the function. Simply wraps
     the function in a with posthog.new_context(): block.
 
     Args:
         fresh: Whether to start with a fresh context (default: False)
+        capture_exceptions: Whether to capture and track exceptions with posthog error tracking (default: True)
 
     Example:
         @posthog.scoped()
@@ -114,7 +119,7 @@ def scoped(fresh=False):
 
         @wraps(func)
         def wrapper(*args, **kwargs):
-            with new_context(fresh=fresh):
+            with new_context(fresh=fresh, capture_exceptions=capture_exceptions):
                 return func(*args, **kwargs)
 
         return cast(F, wrapper)

posthog/sentry/posthog_integration.py

@@ -4,7 +4,7 @@ from sentry_sdk.integrations import Integration
 from sentry_sdk.scope import add_global_event_processor
 from sentry_sdk.utils import Dsn
 
-import posthog
+from posthog import capture, host
 from posthog.request import DEFAULT_HOST
 from posthog.sentry import POSTHOG_ID_TAG
 
@@ -34,7 +34,7 @@ class PostHogIntegration(Integration):
                 if event.get("tags", {}).get(POSTHOG_ID_TAG):
                     posthog_distinct_id = event["tags"][POSTHOG_ID_TAG]
                     event["tags"]["PostHog URL"] = (
-                        f"{posthog.host or DEFAULT_HOST}/person/{posthog_distinct_id}"
+                        f"{host or DEFAULT_HOST}/person/{posthog_distinct_id}"
                     )
 
                     properties = {
@@ -52,6 +52,6 @@ class PostHogIntegration(Integration):
                                 f"{PostHogIntegration.prefix}{PostHogIntegration.organization}/issues/?project={project_id}&query={event['event_id']}"
                             )
 
-                    posthog.capture(posthog_distinct_id, "$exception", properties)
+                    capture(posthog_distinct_id, "$exception", properties)
 
             return event

posthog/test/ai/langchain/test_callbacks.py

@@ -1378,11 +1378,11 @@ def test_langgraph_agent(mock_client):
     )
     graph.invoke(inputs, config={"callbacks": [cb]})
     calls = [call[1] for call in mock_client.capture.call_args_list]
-    assert len(calls) == 21
+    assert len(calls) == 15
     for call in calls:
         assert call["properties"]["$ai_trace_id"] == "test-trace-id"
     assert len([call for call in calls if call["event"] == "$ai_generation"]) == 2
-    assert len([call for call in calls if call["event"] == "$ai_span"]) == 18
+    assert len([call for call in calls if call["event"] == "$ai_span"]) == 12
     assert len([call for call in calls if call["event"] == "$ai_trace"]) == 1
 
 
@@ -1435,11 +1435,13 @@ def test_span_set_parent_ids_for_third_level_run(mock_client, trace_id):
 
     assert mock_client.capture.call_count == 3
 
-    span2, span1, trace = [
-        call[1]["properties"] for call in mock_client.capture.call_args_list
-    ]
-    assert span2["$ai_parent_id"] == span1["$ai_span_id"]
-    assert span1["$ai_parent_id"] == trace["$ai_trace_id"]
+    calls = mock_client.capture.call_args_list
+    span_props_2 = calls[0][1]["properties"]
+    span_props_1 = calls[1][1]["properties"]
+    trace_props = calls[2][1]["properties"]
+
+    assert span_props_2["$ai_parent_id"] == span_props_1["$ai_span_id"]
+    assert span_props_1["$ai_parent_id"] == trace_props["$ai_trace_id"]
 
 
 def test_captures_error_with_details_in_span(mock_client):
@@ -1478,3 +1480,250 @@ def test_captures_error_without_details_in_span(mock_client):
         == "ValueError"
     )
     assert mock_client.capture.call_args_list[1][1]["properties"]["$ai_is_error"]
+
+
+def test_openai_reasoning_tokens(mock_client):
+    """Test that OpenAI reasoning tokens are captured correctly."""
+    prompt = ChatPromptTemplate.from_messages(
+        [("user", "Think step by step about this problem")]
+    )
+
+    # Mock response with reasoning tokens in output_token_details
+    model = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="Let me think through this step by step...",
+                usage_metadata={
+                    "input_tokens": 10,
+                    "output_tokens": 25,
+                    "total_tokens": 35,
+                    "output_token_details": {"reasoning": 15},  # 15 reasoning tokens
+                },
+            )
+        ]
+    )
+
+    callbacks = [CallbackHandler(mock_client)]
+    chain = prompt | model
+    result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "Let me think through this step by step..."
+    assert mock_client.capture.call_count == 3
+
+    generation_args = mock_client.capture.call_args_list[1][1]
+    generation_props = generation_args["properties"]
+
+    assert generation_args["event"] == "$ai_generation"
+    assert generation_props["$ai_input_tokens"] == 10
+    assert generation_props["$ai_output_tokens"] == 25
+    assert generation_props["$ai_reasoning_tokens"] == 15
+
+
+def test_anthropic_cache_write_and_read_tokens(mock_client):
+    """Test that Anthropic cache creation and read tokens are captured correctly."""
+    prompt = ChatPromptTemplate.from_messages([("user", "Analyze this large document")])
+
+    # First call with cache creation
+    model_write = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="I've analyzed the document and cached the context.",
+                usage_metadata={
+                    "total_tokens": 1050,
+                    "input_tokens": 1000,
+                    "output_tokens": 50,
+                    "cache_creation_input_tokens": 800,  # Anthropic cache write
+                },
+            )
+        ]
+    )
+
+    callbacks = [CallbackHandler(mock_client)]
+    chain = prompt | model_write
+    result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "I've analyzed the document and cached the context."
+    assert mock_client.capture.call_count == 3
+
+    generation_args = mock_client.capture.call_args_list[1][1]
+    generation_props = generation_args["properties"]
+
+    assert generation_args["event"] == "$ai_generation"
+    assert generation_props["$ai_input_tokens"] == 1000
+    assert generation_props["$ai_output_tokens"] == 50
+    assert generation_props["$ai_cache_creation_input_tokens"] == 800
+    assert generation_props["$ai_cache_read_input_tokens"] == 0
+    assert generation_props["$ai_reasoning_tokens"] == 0
+
+    # Reset mock for second call
+    mock_client.reset_mock()
+
+    # Second call with cache read
+    model_read = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="Using cached analysis to provide quick response.",
+                usage_metadata={
+                    "input_tokens": 200,
+                    "output_tokens": 30,
+                    "total_tokens": 1030,
+                    "cache_read_input_tokens": 800,  # Anthropic cache read
+                },
+            )
+        ]
+    )
+
+    chain = prompt | model_read
+    result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "Using cached analysis to provide quick response."
+    assert mock_client.capture.call_count == 3
+
+    generation_args = mock_client.capture.call_args_list[1][1]
+    generation_props = generation_args["properties"]
+
+    assert generation_args["event"] == "$ai_generation"
+    assert generation_props["$ai_input_tokens"] == 200
+    assert generation_props["$ai_output_tokens"] == 30
+    assert generation_props["$ai_cache_creation_input_tokens"] == 0
+    assert generation_props["$ai_cache_read_input_tokens"] == 800
+    assert generation_props["$ai_reasoning_tokens"] == 0
+
+
+def test_openai_cache_read_tokens(mock_client):
+    """Test that OpenAI cache read tokens are captured correctly."""
+    prompt = ChatPromptTemplate.from_messages(
+        [("user", "Use the cached prompt for this request")]
+    )
+
+    # Mock response with cache read tokens in input_token_details
+    model = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="Response using cached prompt context.",
+                usage_metadata={
+                    "input_tokens": 150,
+                    "output_tokens": 40,
+                    "total_tokens": 190,
+                    "input_token_details": {
+                        "cache_read": 100,  # 100 tokens read from cache
+                        "cache_creation": 0,
+                    },
+                },
+            )
+        ]
+    )
+
+    callbacks = [CallbackHandler(mock_client)]
+    chain = prompt | model
+    result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "Response using cached prompt context."
+    assert mock_client.capture.call_count == 3
+
+    generation_args = mock_client.capture.call_args_list[1][1]
+    generation_props = generation_args["properties"]
+
+    assert generation_args["event"] == "$ai_generation"
+    assert generation_props["$ai_input_tokens"] == 150
+    assert generation_props["$ai_output_tokens"] == 40
+    assert generation_props["$ai_cache_read_input_tokens"] == 100
+    assert generation_props["$ai_cache_creation_input_tokens"] == 0
+    assert generation_props["$ai_reasoning_tokens"] == 0
+
+
+def test_openai_cache_creation_tokens(mock_client):
+    """Test that OpenAI cache creation tokens are captured correctly."""
+    prompt = ChatPromptTemplate.from_messages(
+        [("user", "Create a cache for this large prompt context")]
+    )
+
+    # Mock response with cache creation tokens in input_token_details
+    model = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="Created cache for the prompt context.",
+                usage_metadata={
+                    "input_tokens": 2000,
+                    "output_tokens": 25,
+                    "total_tokens": 2025,
+                    "input_token_details": {
+                        "cache_creation": 1500,  # 1500 tokens written to cache
+                        "cache_read": 0,
+                    },
+                },
+            )
+        ]
+    )
+
+    callbacks = [CallbackHandler(mock_client)]
+    chain = prompt | model
+    result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "Created cache for the prompt context."
+    assert mock_client.capture.call_count == 3
+
+    generation_args = mock_client.capture.call_args_list[1][1]
+    generation_props = generation_args["properties"]
+
+    assert generation_args["event"] == "$ai_generation"
+    assert generation_props["$ai_input_tokens"] == 2000
+    assert generation_props["$ai_output_tokens"] == 25
+    assert generation_props["$ai_cache_creation_input_tokens"] == 1500
+    assert generation_props["$ai_cache_read_input_tokens"] == 0
+    assert generation_props["$ai_reasoning_tokens"] == 0
+
+
+def test_combined_reasoning_and_cache_tokens(mock_client):
+    """Test that both reasoning tokens and cache tokens can be captured together."""
+    prompt = ChatPromptTemplate.from_messages(
+        [("user", "Think through this cached problem")]
+    )
+
+    # Mock response with both reasoning and cache tokens
+    model = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="Let me reason through this using cached context...",
+                usage_metadata={
+                    "input_tokens": 500,
+                    "output_tokens": 100,
+                    "total_tokens": 600,
+                    "input_token_details": {"cache_read": 300, "cache_creation": 0},
+                    "output_token_details": {"reasoning": 60},  # 60 reasoning tokens
+                },
+            )
+        ]
+    )
+
+    callbacks = [CallbackHandler(mock_client)]
+    chain = prompt | model
+    result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "Let me reason through this using cached context..."
+    assert mock_client.capture.call_count == 3
+
+    generation_args = mock_client.capture.call_args_list[1][1]
+    generation_props = generation_args["properties"]
+
+    assert generation_args["event"] == "$ai_generation"
+    assert generation_props["$ai_input_tokens"] == 500
+    assert generation_props["$ai_output_tokens"] == 100
+    assert generation_props["$ai_cache_read_input_tokens"] == 300
+    assert generation_props["$ai_cache_creation_input_tokens"] == 0
+    assert generation_props["$ai_reasoning_tokens"] == 60
+
+
+@pytest.mark.skipif(not OPENAI_API_KEY, reason="OPENAI_API_KEY is not set")
+def test_openai_reasoning_tokens(mock_client):
+    model = ChatOpenAI(
+        api_key=OPENAI_API_KEY, model="o4-mini", max_completion_tokens=10
+    )
+    cb = CallbackHandler(
+        mock_client, trace_id="test-trace-id", distinct_id="test-distinct-id"
+    )
+    model.invoke("what is the weather in sf", config={"callbacks": [cb]})
+    call = mock_client.capture.call_args_list[0][1]
+    assert call["properties"]["$ai_reasoning_tokens"] is not None
+    assert call["properties"]["$ai_input_tokens"] is not None
+    assert call["properties"]["$ai_output_tokens"] is not None

posthog/test/ai/openai/test_openai.py

@@ -26,6 +26,11 @@ try:
         ResponseOutputMessage,
         ResponseOutputText,
         ResponseUsage,
+        ParsedResponse,
+    )
+    from openai.types.responses.parsed_response import (
+        ParsedResponseOutputMessage,
+        ParsedResponseOutputText,
     )
 
     from posthog.ai.openai import OpenAI
@@ -115,6 +120,59 @@ def mock_openai_response_with_responses_api():
     )
 
 
+@pytest.fixture
+def mock_parsed_response():
+    return ParsedResponse(
+        id="test",
+        model="gpt-4o-2024-08-06",
+        object="response",
+        created_at=1741476542,
+        status="completed",
+        error=None,
+        incomplete_details=None,
+        instructions=None,
+        max_output_tokens=None,
+        tools=[],
+        tool_choice="auto",
+        output=[
+            ParsedResponseOutputMessage(
+                id="msg_123",
+                type="message",
+                role="assistant",
+                status="completed",
+                content=[
+                    ParsedResponseOutputText(
+                        type="output_text",
+                        text='{"name": "Science Fair", "date": "Friday", "participants": ["Alice", "Bob"]}',
+                        annotations=[],
+                        parsed={
+                            "name": "Science Fair",
+                            "date": "Friday",
+                            "participants": ["Alice", "Bob"],
+                        },
+                    )
+                ],
+            )
+        ],
+        output_parsed={
+            "name": "Science Fair",
+            "date": "Friday",
+            "participants": ["Alice", "Bob"],
+        },
+        parallel_tool_calls=True,
+        previous_response_id=None,
+        usage=ResponseUsage(
+            input_tokens=15,
+            output_tokens=20,
+            input_tokens_details={"prompt_tokens": 15, "cached_tokens": 0},
+            output_tokens_details={"reasoning_tokens": 5},
+            total_tokens=35,
+        ),
+        user=None,
+        metadata={},
+    )
+
+
 @pytest.fixture
 def mock_embedding_response():
     return CreateEmbeddingResponse(
@@ -646,3 +704,73 @@ def test_responses_api(mock_client, mock_openai_response_with_responses_api):
         assert props["$ai_http_status"] == 200
         assert props["foo"] == "bar"
         assert isinstance(props["$ai_latency"], float)
+
+
+def test_responses_parse(mock_client, mock_parsed_response):
+    with patch(
+        "openai.resources.responses.Responses.parse",
+        return_value=mock_parsed_response,
+    ):
+        client = OpenAI(api_key="test-key", posthog_client=mock_client)
+        response = client.responses.parse(
+            model="gpt-4o-2024-08-06",
+            input=[
+                {"role": "system", "content": "Extract the event information."},
+                {
+                    "role": "user",
+                    "content": "Alice and Bob are going to a science fair on Friday.",
+                },
+            ],
+            text={
+                "format": {
+                    "type": "json_schema",
+                    "json_schema": {
+                        "name": "event",
+                        "schema": {
+                            "type": "object",
+                            "properties": {
+                                "name": {"type": "string"},
+                                "date": {"type": "string"},
+                                "participants": {
+                                    "type": "array",
+                                    "items": {"type": "string"},
+                                },
+                            },
+                            "required": ["name", "date", "participants"],
+                        },
+                    },
+                }
+            },
+            posthog_distinct_id="test-id",
+            posthog_properties={"foo": "bar"},
+        )
+
+        assert response == mock_parsed_response
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+
+        assert call_args["distinct_id"] == "test-id"
+        assert call_args["event"] == "$ai_generation"
+        assert props["$ai_provider"] == "openai"
+        assert props["$ai_model"] == "gpt-4o-2024-08-06"
+        assert props["$ai_input"] == [
+            {"role": "system", "content": "Extract the event information."},
+            {
+                "role": "user",
+                "content": "Alice and Bob are going to a science fair on Friday.",
+            },
+        ]
+        assert props["$ai_output_choices"] == [
+            {
+                "role": "assistant",
+                "content": '{"name": "Science Fair", "date": "Friday", "participants": ["Alice", "Bob"]}',
+            }
+        ]
+        assert props["$ai_input_tokens"] == 15
+        assert props["$ai_output_tokens"] == 20
+        assert props["$ai_reasoning_tokens"] == 5
+        assert props["$ai_http_status"] == 200
+        assert props["foo"] == "bar"
+        assert isinstance(props["$ai_latency"], float)

posthog/test/exception_integrations/test_django.py

@@ -1,20 +1,44 @@
 from posthog.exception_integrations.django import DjangoRequestExtractor
+from django.test import RequestFactory
+from django.conf import settings
+from django.core.management import call_command
+import django
 
 DEFAULT_USER_AGENT = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/58.0.3029.110 Safari/537.3"
 
+# setup a test app
+if not settings.configured:
+    settings.configure(
+        SECRET_KEY="test",
+        DEFAULT_CHARSET="utf-8",
+        INSTALLED_APPS=[
+            "django.contrib.auth",
+            "django.contrib.contenttypes",
+        ],
+        DATABASES={
+            "default": {
+                "ENGINE": "django.db.backends.sqlite3",
+                "NAME": ":memory:",
+            }
+        },
+    )
+    django.setup()
+
+    call_command("migrate", verbosity=0, interactive=False)
+
 
 def mock_request_factory(override_headers):
-    class Request:
-        META = {}
-        # TRICKY: Actual django request dict object has case insensitive matching, and strips http from the names
-        headers = {
+    factory = RequestFactory(
+        headers={
             "User-Agent": DEFAULT_USER_AGENT,
             "Referrer": "http://example.com",
             "X-Forwarded-For": "193.4.5.12",
             **(override_headers or {}),
         }
+    )
 
-    return Request()
+    request = factory.get("/api/endpoint")
+    return request
 
 
 def test_request_extractor_with_no_trace():
@@ -25,6 +49,7 @@ def test_request_extractor_with_no_trace():
         "user_agent": DEFAULT_USER_AGENT,
         "traceparent": None,
         "distinct_id": None,
+        "$request_path": "/api/endpoint",
     }
 
 
@@ -32,12 +57,14 @@ def test_request_extractor_with_trace():
     request = mock_request_factory(
         {"traceparent": "00-0af7651916cd43dd8448eb211c80319c-b7ad6b7169203331-01"}
     )
+
     extractor = DjangoRequestExtractor(request)
     assert extractor.extract_person_data() == {
         "ip": "193.4.5.12",
         "user_agent": DEFAULT_USER_AGENT,
         "traceparent": "00-0af7651916cd43dd8448eb211c80319c-b7ad6b7169203331-01",
         "distinct_id": None,
+        "$request_path": "/api/endpoint",
     }
 
 
@@ -54,6 +81,7 @@ def test_request_extractor_with_tracestate():
         "user_agent": DEFAULT_USER_AGENT,
         "traceparent": "00-0af7651916cd43dd8448eb211c80319c-b7ad6b7169203331-01",
         "distinct_id": "1234",
+        "$request_path": "/api/endpoint",
     }
 
 
@@ -67,4 +95,27 @@ def test_request_extractor_with_complicated_tracestate():
         "user_agent": DEFAULT_USER_AGENT,
         "traceparent": None,
         "distinct_id": "alohaMountainsXUYZ",
+        "$request_path": "/api/endpoint",
+    }
+
+
+def test_request_extractor_with_request_user():
+    from django.contrib.auth.models import User
+
+    user = User.objects.create_user(
+        username="test", email="test@posthog.com", password="top_secret"
+    )
+
+    request = mock_request_factory(None)
+    request.user = user
+
+    extractor = DjangoRequestExtractor(request)
+    assert extractor.extract_person_data() == {
+        "ip": "193.4.5.12",
+        "user_agent": DEFAULT_USER_AGENT,
+        "traceparent": None,
+        "distinct_id": None,
+        "$request_path": "/api/endpoint",
+        "email": "test@posthog.com",
+        "$user_id": "1",
     }

posthog/test/test_before_send.py

@@ -0,0 +1,171 @@
+import unittest
+
+import mock
+
+from posthog.client import Client
+from posthog.test.test_utils import FAKE_TEST_API_KEY
+
+
+class TestClient(unittest.TestCase):
+    @classmethod
+    def setUpClass(cls):
+        # This ensures no real HTTP POST requests are made
+        cls.client_post_patcher = mock.patch("posthog.client.batch_post")
+        cls.consumer_post_patcher = mock.patch("posthog.consumer.batch_post")
+        cls.client_post_patcher.start()
+        cls.consumer_post_patcher.start()
+
+    @classmethod
+    def tearDownClass(cls):
+        cls.client_post_patcher.stop()
+        cls.consumer_post_patcher.stop()
+
+    def set_fail(self, e, batch):
+        """Mark the failure handler"""
+        print("FAIL", e, batch)  # noqa: T201
+        self.failed = True
+
+    def setUp(self):
+        self.failed = False
+        self.client = Client(FAKE_TEST_API_KEY, on_error=self.set_fail)
+
+    def test_before_send_callback_modifies_event(self):
+        """Test that before_send callback can modify events."""
+        processed_events = []
+
+        def my_before_send(event):
+            processed_events.append(event.copy())
+            if "properties" not in event:
+                event["properties"] = {}
+            event["properties"]["processed_by_before_send"] = True
+            return event
+
+        client = Client(
+            FAKE_TEST_API_KEY, on_error=self.set_fail, before_send=my_before_send
+        )
+        success, msg = client.capture("user1", "test_event", {"original": "value"})
+
+        self.assertTrue(success)
+        self.assertEqual(msg["properties"]["processed_by_before_send"], True)
+        self.assertEqual(msg["properties"]["original"], "value")
+        self.assertEqual(len(processed_events), 1)
+        self.assertEqual(processed_events[0]["event"], "test_event")
+
+    def test_before_send_callback_drops_event(self):
+        """Test that before_send callback can drop events by returning None."""
+
+        def drop_test_events(event):
+            if event.get("event") == "test_drop_me":
+                return None
+            return event
+
+        client = Client(
+            FAKE_TEST_API_KEY, on_error=self.set_fail, before_send=drop_test_events
+        )
+
+        # Event should be dropped
+        success, msg = client.capture("user1", "test_drop_me")
+        self.assertTrue(success)
+        self.assertIsNone(msg)
+
+        # Event should go through
+        success, msg = client.capture("user1", "keep_me")
+        self.assertTrue(success)
+        self.assertIsNotNone(msg)
+        self.assertEqual(msg["event"], "keep_me")
+
+    def test_before_send_callback_handles_exceptions(self):
+        """Test that exceptions in before_send don't crash the client."""
+
+        def buggy_before_send(event):
+            raise ValueError("Oops!")
+
+        client = Client(
+            FAKE_TEST_API_KEY, on_error=self.set_fail, before_send=buggy_before_send
+        )
+        success, msg = client.capture("user1", "robust_event")
+
+        # Event should still be sent despite the exception
+        self.assertTrue(success)
+        self.assertIsNotNone(msg)
+        self.assertEqual(msg["event"], "robust_event")
+
+    def test_before_send_callback_works_with_all_event_types(self):
+        """Test that before_send works with capture, identify, set, etc."""
+
+        def add_marker(event):
+            if "properties" not in event:
+                event["properties"] = {}
+            event["properties"]["marked"] = True
+            return event
+
+        client = Client(
+            FAKE_TEST_API_KEY, on_error=self.set_fail, before_send=add_marker
+        )
+
+        # Test capture
+        success, msg = client.capture("user1", "event")
+        self.assertTrue(success)
+        self.assertTrue(msg["properties"]["marked"])
+
+        # Test identify
+        success, msg = client.identify("user1", {"trait": "value"})
+        self.assertTrue(success)
+        self.assertTrue(msg["properties"]["marked"])
+
+        # Test set
+        success, msg = client.set("user1", {"prop": "value"})
+        self.assertTrue(success)
+        self.assertTrue(msg["properties"]["marked"])
+
+        # Test page
+        success, msg = client.page("user1", "https://example.com")
+        self.assertTrue(success)
+        self.assertTrue(msg["properties"]["marked"])
+
+    def test_before_send_callback_disabled_when_none(self):
+        """Test that client works normally when before_send is None."""
+        client = Client(FAKE_TEST_API_KEY, on_error=self.set_fail, before_send=None)
+        success, msg = client.capture("user1", "normal_event")
+
+        self.assertTrue(success)
+        self.assertIsNotNone(msg)
+        self.assertEqual(msg["event"], "normal_event")
+
+    def test_before_send_callback_pii_scrubbing_example(self):
+        """Test a realistic PII scrubbing use case."""
+
+        def scrub_pii(event):
+            properties = event.get("properties", {})
+
+            # Mask email but keep domain
+            if "email" in properties:
+                email = properties["email"]
+                if "@" in email:
+                    domain = email.split("@")[1]
+                    properties["email"] = f"***@{domain}"
+                else:
+                    properties["email"] = "***"
+
+            # Remove credit card
+            properties.pop("credit_card", None)
+
+            return event
+
+        client = Client(
+            FAKE_TEST_API_KEY, on_error=self.set_fail, before_send=scrub_pii
+        )
+        success, msg = client.capture(
+            "user1",
+            "form_submit",
+            {
+                "email": "user@example.com",
+                "credit_card": "1234-5678-9012-3456",
+                "form_name": "contact",
+            },
+        )
+
+        self.assertTrue(success)
+        self.assertEqual(msg["properties"]["email"], "***@example.com")
+        self.assertNotIn("credit_card", msg["properties"])
+        self.assertEqual(msg["properties"]["form_name"], "contact")

posthog/test/test_client.py

@@ -1,4 +1,3 @@
-import hashlib
 import time
 import unittest
 from datetime import datetime
@@ -8,7 +7,7 @@ import mock
 import six
 from parameterized import parameterized
 
-from posthog.client import EXCLUDED_HASHES, INCLUDED_HASHES, Client, is_token_in_rollout
+from posthog.client import Client
 from posthog.request import APIError
 from posthog.test.test_utils import FAKE_TEST_API_KEY
 from posthog.types import FeatureFlag, LegacyFlagMetadata
@@ -1462,80 +1461,3 @@ class TestClient(unittest.TestCase):
             "errorsWhileComputingFlags": False,
             "requestId": "test-id",
         }
-
-    @mock.patch("posthog.client.decide")
-    @mock.patch("posthog.client.flags")
-    def test_get_flags_decision_rollout(self, patch_flags, patch_decide):
-        # Set up mock responses
-        decide_response = {
-            "featureFlags": {"flag1": True},
-            "featureFlagPayloads": {},
-            "errorsWhileComputingFlags": False,
-        }
-        flags_response = {
-            "featureFlags": {"flag2": True},
-            "featureFlagPayloads": {},
-            "errorsWhileComputingFlags": False,
-        }
-        patch_decide.return_value = decide_response
-        patch_flags.return_value = flags_response
-
-        client = Client(FAKE_TEST_API_KEY)
-
-        # Test 100% rollout - should use flags
-        with mock.patch(
-            "posthog.client.is_token_in_rollout", return_value=True
-        ) as mock_rollout:
-            client.get_flags_decision("distinct_id")
-            mock_rollout.assert_called_with(
-                FAKE_TEST_API_KEY,
-                1,
-                included_hashes=INCLUDED_HASHES,
-                excluded_hashes=EXCLUDED_HASHES,
-            )
-            patch_flags.assert_called_once()
-            patch_decide.assert_not_called()
-
-    def test_token_rollout_calculation(self):
-        # Test specific hash inclusion
-        token = "test_token"
-        token_hash = hashlib.sha1(token.encode("utf-8")).hexdigest()
-        included_hashes = {token_hash}
-
-        # Should be included due to specific hash, even with 0% rollout
-        self.assertTrue(
-            expr=is_token_in_rollout(
-                token, percentage=0.0, included_hashes=included_hashes
-            )
-        )
-
-        # Should not be included with 0% rollout and no specific hash
-        self.assertFalse(is_token_in_rollout(token, percentage=0.0))
-
-        # Should be included with 100% rollout regardless of specific hash
-        self.assertTrue(is_token_in_rollout(token, percentage=1.0))
-        self.assertTrue(
-            is_token_in_rollout(token, percentage=1.0, included_hashes=included_hashes)
-        )
-
-        # Test deterministic behavior - same token should always give same result
-        hash_float = int(token_hash[:8], 16) / 0xFFFFFFFF
-        percentage = hash_float + 0.1  # Just above the hash value
-
-        self.assertTrue(is_token_in_rollout(token, percentage))
-        self.assertFalse(
-            is_token_in_rollout(token, percentage - 0.2)
-        )  # Just below the hash value
-
-        # Test that the token exclusion works correctly
-        self.assertFalse(
-            is_token_in_rollout(token, percentage=1.0, excluded_hashes={token_hash})
-        )
-
-        # Should work for other specific token hashes
-        # Include our API key
-        self.assertTrue(
-            is_token_in_rollout(
-                "sTMFPsFhdP1Ssg", percentage=0.1, included_hashes=INCLUDED_HASHES
-            )
-        )

posthog/types.py

@@ -1,9 +1,13 @@
 import json
 from dataclasses import dataclass
-from typing import Any, List, Optional, TypedDict, Union, cast
+from typing import Any, Callable, List, Optional, TypedDict, Union, cast
 
 FlagValue = Union[bool, str]
 
+# Type alias for the before_send callback function
+# Takes an event dictionary and returns the modified event or None to drop it
+BeforeSendCallback = Callable[[dict[str, Any]], Optional[dict[str, Any]]]
+
 
 @dataclass(frozen=True)
 class FlagReason:

posthog/version.py

@@ -1,4 +1,4 @@
-VERSION = "4.3.3"
+VERSION = "4.9.0"
 
 if __name__ == "__main__":
     print(VERSION, end="")  # noqa: T201

pyproject.toml

@@ -8,13 +8,14 @@ dynamic = ["version"]
 description = "Integrate PostHog into any python application."
 authors = [{ name = "PostHog", email = "hey@posthog.com" }]
 maintainers = [{ name = "PostHog", email = "hey@posthog.com" }]
-license = { text = "MIT License" }
+license = {text = "MIT"}
 readme = "README.md"
 requires-python = ">=3.9"
 classifiers = [
     "Development Status :: 5 - Production/Stable",
     "Intended Audience :: Developers",
     "Operating System :: OS Independent",
+    "License :: OSI Approved :: MIT License",
     "Programming Language :: Python",
     "Programming Language :: Python :: 3.9",
     "Programming Language :: Python :: 3.10",
@@ -35,6 +36,8 @@ Homepage = "https://github.com/posthog/posthog-python"
 Repository = "https://github.com/posthog/posthog-python"
 
 [project.optional-dependencies]
+sentry = ["sentry-sdk", "django"]
+langchain = ["langchain>=0.2.0"]
 dev = [
     "django-stubs",
     "lxml",
@@ -48,6 +51,12 @@ dev = [
     "pre-commit",
     "pydantic",
     "ruff",
+    "setuptools",
+    "packaging", 
+    "wheel",
+    "twine",
+    "tomli",
+    "tomli_w",
 ]
 test = [
     "mock>=2.0.0",
@@ -59,16 +68,15 @@ test = [
     "django",
     "openai",
     "anthropic",
-    "langgraph",
-    "langchain-community>=0.2.0",
-    "langchain-openai>=0.2.0",
-    "langchain-anthropic>=0.2.0",
+    "langgraph>=0.4.8",
+    "langchain-core>=0.3.65",
+    "langchain-community>=0.3.25",
+    "langchain-openai>=0.3.22",
+    "langchain-anthropic>=0.3.15",
     "google-genai",
     "pydantic",
     "parameterized>=0.8.1",
 ]
-sentry = ["sentry-sdk", "django"]
-langchain = ["langchain>=0.2.0"]
 
 [tool.setuptools]
 packages = [
@@ -83,8 +91,6 @@ packages = [
     "posthog.exception_integrations",
 ]
 
-license-files = []
-
 [tool.setuptools.dynamic]
 version = { attr = "posthog.version.VERSION" }
 

setup.cfg

@@ -1,12 +0,0 @@
-[bdist_wheel]
-universal = 1
-
-[tool:pytest]
-asyncio_mode = auto
-asyncio_default_fixture_loop_scope = function
-
-[flake8]
-# ignore E501 for line length
-# ignore W503 for line break before binary operator
-ignore = E501,W503
-max-line-length = 120

setup.py

@@ -28,7 +28,6 @@ setup(
     author_email="hey@posthog.com",
     maintainer="PostHog",
     maintainer_email="hey@posthog.com",
-    test_suite="posthog.test.all",
     license="MIT License",
     description="Integrate PostHog into any python application.",
     long_description=long_description,

setup_analytics.py

@@ -1,5 +1,8 @@
 import os
 import sys
+import tomli
+import tomli_w
+import shutil
 
 try:
     from setuptools import setup
@@ -7,9 +10,39 @@ except ImportError:
     from distutils.core import setup
 
 # Don't import analytics-python module here, since deps may not be installed
-sys.path.insert(0, os.path.join(os.path.dirname(__file__), "posthog"))
+sys.path.insert(0, os.path.join(os.path.dirname(__file__), "posthoganalytics"))
 from version import VERSION  # noqa: E402
 
+
+# Copy the original pyproject.toml as backup
+shutil.copy("pyproject.toml", "pyproject.toml.backup")
+
+# Read the original pyproject.toml
+with open("pyproject.toml", "rb") as f:
+    config = tomli.load(f)
+
+# Override specific values
+config["project"]["name"] = "posthoganalytics"
+config["tool"]["setuptools"]["dynamic"]["version"] = {
+    "attr": "posthoganalytics.version.VERSION"
+}
+
+# Rename packages from posthog.* to posthoganalytics.*
+if "packages" in config["tool"]["setuptools"]:
+    new_packages = []
+    for package in config["tool"]["setuptools"]["packages"]:
+        if package == "posthog":
+            new_packages.append("posthoganalytics")
+        elif package.startswith("posthog."):
+            new_packages.append(package.replace("posthog.", "posthoganalytics.", 1))
+        else:
+            new_packages.append(package)
+    config["tool"]["setuptools"]["packages"] = new_packages
+
+# Overwrite the original pyproject.toml
+with open("pyproject.toml", "wb") as f:
+    tomli_w.dump(config, f)
+
 long_description = """
 PostHog is developer-friendly, self-hosted product analytics.
 posthog-python is the python package.
@@ -28,7 +61,6 @@ setup(
     author_email="hey@posthog.com",
     maintainer="PostHog",
     maintainer_email="hey@posthog.com",
-    test_suite="posthog.test.all",
     license="MIT License",
     description="Integrate PostHog into any python application.",
     long_description=long_description,