fix(llm-observability): capture system prompt for anthropic (#167)

Co-authored-by: Peter Kirkham <peter@posthog.com>

.github/workflows/ci.yml

@@ -18,7 +18,7 @@ jobs:
               with:
                   python-version: 3.8
 
-            - uses: actions/cache@v1
+            - uses: actions/cache@v3
               with:
                   path: ~/.cache/pip
                   key: ${{ runner.os }}-pip-${{ hashFiles('setup.py') }}
@@ -33,7 +33,7 @@ jobs:
             - name: Check formatting with black
               run: |
                   black --check .
-            
+
             - name: Lint with flake8
               run: |
                   flake8 posthog --ignore E501
@@ -47,14 +47,14 @@ jobs:
         runs-on: ubuntu-latest
 
         steps:
-            - uses: actions/checkout@v1
+            - uses: actions/checkout@v2
               with:
                   fetch-depth: 1
 
-            - name: Set up Python 3.7
-              uses: actions/setup-python@v1
+            - name: Set up Python 3.9
+              uses: actions/setup-python@v2
               with:
-                  python-version: 3.7
+                  python-version: 3.9
 
             - name: Install requirements.txt dependencies with pip
               run: |

.gitignore

@@ -14,4 +14,6 @@ pylint.out
 posthog-analytics
 .idea
 .python-version
-.coverage
+.coverage
+pyrightconfig.json
+.env

CHANGELOG.md

@@ -1,8 +1,112 @@
+## 3.8.4 - 2025-01-17
+
+1. Add Anthropic support for LLM Observability.
+2. Update LLM Observability to use output_choices.
+
+## 3.8.3 - 2025-01-14
+
+1. Fix setuptools to include the `posthog.ai.openai` and `posthog.ai.langchain` packages for the `posthoganalytics` package.
+
+## 3.8.2 - 2025-01-14
+
+1. Fix setuptools to include the `posthog.ai.openai` and `posthog.ai.langchain` packages.
+
+## 3.8.1 - 2025-01-14
+
+1. Add LLM Observability with support for OpenAI and Langchain callbacks.
+
+## 3.7.5 - 2025-01-03
+
+1. Add `distinct_id` to group_identify
+
+## 3.7.4 - 2024-11-25
+
+1. Fix bug where this SDK incorrectly sent feature flag events with null values when calling `get_feature_flag_payload`.
+
+## 3.7.3 - 2024-11-25
+
+1. Use personless mode when sending an exception without a provided `distinct_id`.
+
+## 3.7.2 - 2024-11-19
+
+1. Add `type` property to exception stacks.
+
+## 3.7.1 - 2024-10-24
+
+1. Add `platform` property to each frame of exception stacks.
+
+## 3.7.0 - 2024-10-03
+
+1. Adds a new `super_properties` parameter on the client that are appended to every /capture call.
+
+## 3.6.7 - 2024-09-24
+
+1. Remove deprecated datetime.utcnow() in favour of datetime.now(tz=tzutc())
+
+## 3.6.6 - 2024-09-16
+
+1. Fix manual capture support for in app frames
+
+## 3.6.5 - 2024-09-10
+
+1. Fix django integration support for manual exception capture.
+
+## 3.6.4 - 2024-09-05
+
+1. Add manual exception capture.
+
+## 3.6.3 - 2024-09-03
+
+1. Make sure setup.py for posthoganalytics package also discovers the new exception integration package.
+
+## 3.6.2 - 2024-09-03
+
+1. Make sure setup.py discovers the new exception integration package.
+
+## 3.6.1 - 2024-09-03
+
+1. Adds django integration to exception autocapture in alpha state. This feature is not yet stable and may change in future versions.
+
+## 3.6.0 - 2024-08-28
+
+1. Adds exception autocapture in alpha state. This feature is not yet stable and may change in future versions.
+
+## 3.5.2 - 2024-08-21
+
+1. Guard for None values in local evaluation
+
+## 3.5.1 - 2024-08-13
+
+1. Remove "-api" suffix from ingestion hostnames
+
+## 3.5.0 - 2024-02-29
+
+1. - Adds a new `feature_flags_request_timeout_seconds` timeout parameter for feature flags which defaults to 3 seconds, updated from the default 10s for all other API calls.
+
+## 3.4.2 - 2024-02-20
+
+1. Add `historical_migration` option for bulk migration to PostHog Cloud.
+
+## 3.4.1 - 2024-02-09
+
+1. Use new hosts for event capture as well
+
+## 3.4.0 - 2024-02-05
+
+1. Point given hosts to new ingestion hosts
+
+## 3.3.4 - 2024-01-30
+
+1. Update type hints for module variables to work with newer versions of mypy
+
+## 3.3.3 - 2024-01-26
+
+1. Remove new relative date operators, combine into regular date operators
+
 ## 3.3.2 - 2024-01-19
 
 1. Return success/failure with all capture calls from module functions
 
-
 ## 3.3.1 - 2024-01-10
 
 1. Make sure we don't override any existing feature flag properties when adding locally evaluated feature flag properties.
@@ -28,6 +132,7 @@
 
 1. Restore how feature flags work when the client library is disabled: All requests return `None` and no events are sent when the client is disabled.
 2. Add a `feature_flag_definitions()` debug option, which returns currently loaded feature flag definitions. You can use this to more cleverly decide when to request local evaluation of feature flags.
+
 ## 3.0.0 - 2023-04-14
 
 Breaking change:
@@ -68,7 +173,6 @@ posthog = Posthog('api_key', disable_geoip=False)
 1. Log instead of raise error on posthog personal api key errors
 2. Remove upper bound on backoff dependency
 
-
 ## 2.3.0 - 2023-01-31
 
 1. Add support for returning payloads of matched feature flags
@@ -84,6 +188,7 @@ Changes:
 Changes:
 
 1. Fixes issues with date comparison.
+
 ## 2.1.1 - 2022-09-14
 
 Changes:
@@ -96,6 +201,7 @@ Changes:
 
 1. Feature flag defaults have been removed
 2. Setup logging only when debug mode is enabled.
+
 ## 2.0.1 - 2022-08-04
 
 - Make poll_interval configurable
@@ -107,8 +213,8 @@ Changes:
 Breaking changes:
 
 1. The minimum version requirement for PostHog servers is now 1.38. If you're using PostHog Cloud, you satisfy this requirement automatically.
-2. Feature flag defaults apply only when there's an error fetching feature flag results. Earlier, if the default was set to `True`, even if a flag resolved to `False`, the default would override this. 
-**Note: These are removed in 2.0.2**
+2. Feature flag defaults apply only when there's an error fetching feature flag results. Earlier, if the default was set to `True`, even if a flag resolved to `False`, the default would override this.
+   **Note: These are removed in 2.0.2**
 3. Feature flag remote evaluation doesn't require a personal API key.
 
 New Changes:
@@ -116,18 +222,20 @@ New Changes:
 1. You can now evaluate feature flags locally (i.e. without sending a request to your PostHog servers) by setting a personal API key, and passing in groups and person properties to `is_feature_enabled` and `get_feature_flag` calls.
 2. Introduces a `get_all_flags` method that returns all feature flags. This is useful for when you want to seed your frontend with some initial flags, given a user ID.
 
-
-
 ## 1.4.9 - 2022-06-13
+
 - Support for sending feature flags with capture calls
 
 ## 1.4.8 - 2022-05-12
+
 - Support multi variate feature flags
 
 ## 1.4.7 - 2022-04-25
+
 - Allow feature flags usage without project_api_key
 
 ## 1.4.1 - 2021-05-28
+
 - Fix packaging issues with Sentry integrations
 
 ## 1.4.0 - 2021-05-18

CODEOWNERS

@@ -0,0 +1 @@
+@PostHog/team-feature-flags

example.py

@@ -1,7 +1,7 @@
 # PostHog Python library example
 
 # Import the library
-import time
+# import time
 
 import posthog
 

llm_observability_examples.py

@@ -0,0 +1,196 @@
+import os
+import uuid
+
+import posthog
+from posthog.ai.openai import AsyncOpenAI, OpenAI
+
+# Example credentials - replace these with your own or use environment variables
+posthog.project_api_key = os.getenv("POSTHOG_PROJECT_API_KEY", "your-project-api-key")
+posthog.personal_api_key = os.getenv("POSTHOG_PERSONAL_API_KEY", "your-personal-api-key")
+posthog.host = os.getenv("POSTHOG_HOST", "http://localhost:8000")  # Or https://app.posthog.com
+posthog.debug = True
+# change this to False to see usage events
+# posthog.privacy_mode = True
+
+openai_client = OpenAI(
+    api_key=os.getenv("OPENAI_API_KEY", "your-openai-api-key"),
+    posthog_client=posthog,
+)
+
+async_openai_client = AsyncOpenAI(
+    api_key=os.getenv("OPENAI_API_KEY", "your-openai-api-key"),
+    posthog_client=posthog,
+)
+
+
+def main_sync():
+    trace_id = str(uuid.uuid4())
+    print("Trace ID:", trace_id)
+    distinct_id = "test2_distinct_id"
+    properties = {"test_property": "test_value"}
+    groups = {"company": "test_company"}
+
+    try:
+        basic_openai_call(distinct_id, trace_id, properties, groups)
+        streaming_openai_call(distinct_id, trace_id, properties, groups)
+        embedding_openai_call(distinct_id, trace_id, properties, groups)
+        image_openai_call()
+    except Exception as e:
+        print("Error during OpenAI call:", str(e))
+
+
+async def main_async():
+    trace_id = str(uuid.uuid4())
+    print("Trace ID:", trace_id)
+    distinct_id = "test_distinct_id"
+    properties = {"test_property": "test_value"}
+    groups = {"company": "test_company"}
+
+    try:
+        await basic_async_openai_call(distinct_id, trace_id, properties, groups)
+        await streaming_async_openai_call(distinct_id, trace_id, properties, groups)
+        await embedding_async_openai_call(distinct_id, trace_id, properties, groups)
+        await image_async_openai_call()
+    except Exception as e:
+        print("Error during OpenAI call:", str(e))
+
+
+def basic_openai_call(distinct_id, trace_id, properties, groups):
+    response = openai_client.chat.completions.create(
+        model="gpt-4o-mini",
+        messages=[
+            {"role": "system", "content": "You are a complex problem solver."},
+            {"role": "user", "content": "Explain quantum computing in simple terms."},
+        ],
+        max_tokens=100,
+        temperature=0.7,
+        posthog_distinct_id=distinct_id,
+        posthog_trace_id=trace_id,
+        posthog_properties=properties,
+        posthog_groups=groups,
+    )
+    print(response)
+    if response and response.choices:
+        print("OpenAI response:", response.choices[0].message.content)
+    else:
+        print("No response or unexpected format returned.")
+    return response
+
+
+async def basic_async_openai_call(distinct_id, trace_id, properties, groups):
+    response = await async_openai_client.chat.completions.create(
+        model="gpt-4o-mini",
+        messages=[
+            {"role": "system", "content": "You are a complex problem solver."},
+            {"role": "user", "content": "Explain quantum computing in simple terms."},
+        ],
+        max_tokens=100,
+        temperature=0.7,
+        posthog_distinct_id=distinct_id,
+        posthog_trace_id=trace_id,
+        posthog_properties=properties,
+        posthog_groups=groups,
+    )
+    if response and hasattr(response, "choices"):
+        print("OpenAI response:", response.choices[0].message.content)
+    else:
+        print("No response or unexpected format returned.")
+    return response
+
+
+def streaming_openai_call(distinct_id, trace_id, properties, groups):
+
+    response = openai_client.chat.completions.create(
+        model="gpt-4o-mini",
+        messages=[
+            {"role": "system", "content": "You are a complex problem solver."},
+            {"role": "user", "content": "Explain quantum computing in simple terms."},
+        ],
+        max_tokens=100,
+        temperature=0.7,
+        stream=True,
+        posthog_distinct_id=distinct_id,
+        posthog_trace_id=trace_id,
+        posthog_properties=properties,
+        posthog_groups=groups,
+    )
+
+    for chunk in response:
+        if hasattr(chunk, "choices") and chunk.choices and len(chunk.choices) > 0:
+            print(chunk.choices[0].delta.content or "", end="")
+
+    return response
+
+
+async def streaming_async_openai_call(distinct_id, trace_id, properties, groups):
+    response = await async_openai_client.chat.completions.create(
+        model="gpt-4o-mini",
+        messages=[
+            {"role": "system", "content": "You are a complex problem solver."},
+            {"role": "user", "content": "Explain quantum computing in simple terms."},
+        ],
+        max_tokens=100,
+        temperature=0.7,
+        stream=True,
+        posthog_distinct_id=distinct_id,
+        posthog_trace_id=trace_id,
+        posthog_properties=properties,
+        posthog_groups=groups,
+    )
+
+    async for chunk in response:
+        if hasattr(chunk, "choices") and chunk.choices and len(chunk.choices) > 0:
+            print(chunk.choices[0].delta.content or "", end="")
+
+    return response
+
+
+# none instrumented
+def image_openai_call():
+    response = openai_client.images.generate(model="dall-e-3", prompt="A cute baby hedgehog", n=1, size="1024x1024")
+    print(response)
+    return response
+
+
+# none instrumented
+async def image_async_openai_call():
+    response = await async_openai_client.images.generate(
+        model="dall-e-3", prompt="A cute baby hedgehog", n=1, size="1024x1024"
+    )
+    print(response)
+    return response
+
+
+def embedding_openai_call(posthog_distinct_id, posthog_trace_id, posthog_properties, posthog_groups):
+    response = openai_client.embeddings.create(
+        input="The hedgehog is cute",
+        model="text-embedding-3-small",
+        posthog_distinct_id=posthog_distinct_id,
+        posthog_trace_id=posthog_trace_id,
+        posthog_properties=posthog_properties,
+        posthog_groups=posthog_groups,
+    )
+    print(response)
+    return response
+
+
+async def embedding_async_openai_call(posthog_distinct_id, posthog_trace_id, posthog_properties, posthog_groups):
+    response = await async_openai_client.embeddings.create(
+        input="The hedgehog is cute",
+        model="text-embedding-3-small",
+        posthog_distinct_id=posthog_distinct_id,
+        posthog_trace_id=posthog_trace_id,
+        posthog_properties=posthog_properties,
+        posthog_groups=posthog_groups,
+    )
+    print(response)
+    return response
+
+
+# HOW TO RUN:
+# comment out one of these to run the other
+
+if __name__ == "__main__":
+    main_sync()
+
+# asyncio.run(main_async())

posthog/__init__.py

@@ -1,25 +1,35 @@
 import datetime  # noqa: F401
-from typing import Callable, Dict, Optional, Tuple  # noqa: F401
+from typing import Callable, Dict, List, Optional, Tuple  # noqa: F401
 
 from posthog.client import Client
+from posthog.exception_capture import Integrations  # noqa: F401
 from posthog.version import VERSION
 
 __version__ = VERSION
 
 """Settings."""
-api_key = None  # type: str
-host = None  # type: str
-on_error = None  # type: Callable
+api_key = None  # type: Optional[str]
+host = None  # type: Optional[str]
+on_error = None  # type: Optional[Callable]
 debug = False  # type: bool
 send = True  # type: bool
 sync_mode = False  # type: bool
 disabled = False  # type: bool
-personal_api_key = None  # type: str
-project_api_key = None  # type: str
+personal_api_key = None  # type: Optional[str]
+project_api_key = None  # type: Optional[str]
 poll_interval = 30  # type: int
 disable_geoip = True  # type: bool
+feature_flags_request_timeout_seconds = 3  # type: int
+super_properties = None  # type: Optional[Dict]
+# Currently alpha, use at your own risk
+enable_exception_autocapture = False  # type: bool
+exception_autocapture_integrations = []  # type: List[Integrations]
+# Used to determine in app paths for exception autocapture. Defaults to the current working directory
+project_root = None  # type: Optional[str]
+# Used for our AI observability feature to not capture any prompt or output just usage + metadata
+privacy_mode = False  # type: bool
 
-default_client = None
+default_client = None  # type: Optional[Client]
 
 
 def capture(
@@ -246,6 +256,50 @@ def alias(
     )
 
 
+def capture_exception(
+    exception=None,  # type: Optional[BaseException]
+    distinct_id=None,  # type: Optional[str]
+    properties=None,  # type: Optional[Dict]
+    context=None,  # type: Optional[Dict]
+    timestamp=None,  # type: Optional[datetime.datetime]
+    uuid=None,  # type: Optional[str]
+    groups=None,  # type: Optional[Dict]
+):
+    # type: (...) -> Tuple[bool, dict]
+    """
+    capture_exception allows you to capture exceptions that happen in your code. This is useful for debugging and understanding what errors your users are encountering.
+    This function never raises an exception, even if it fails to send the event.
+
+    A `capture_exception` call does not require any fields, but we recommend sending:
+    - `distinct id` which uniquely identifies your user for which this exception happens
+    - `exception` to specify the exception to capture. If not provided, the current exception is captured via `sys.exc_info()`
+
+    Optionally you can submit
+    - `properties`, which can be a dict with any information you'd like to add
+    - `groups`, which is a dict of group type -> group key mappings
+
+    For example:
+    ```python
+    try:
+        1 / 0
+    except Exception as e:
+        posthog.capture_exception(e, 'my specific distinct id')
+        posthog.capture_exception(distinct_id='my specific distinct id')
+
+    ```
+    """
+    return _proxy(
+        "capture_exception",
+        exception=exception,
+        distinct_id=distinct_id,
+        properties=properties,
+        context=context,
+        timestamp=timestamp,
+        uuid=uuid,
+        groups=groups,
+    )
+
+
 def feature_enabled(
     key,  # type: str
     distinct_id,  # type: str
@@ -405,6 +459,11 @@ def feature_flag_definitions():
     return _proxy("feature_flag_definitions")
 
 
+def load_feature_flags():
+    """Load feature flag definitions from PostHog."""
+    return _proxy("load_feature_flags")
+
+
 def page(*args, **kwargs):
     """Send a page call."""
     _proxy("page", *args, **kwargs)
@@ -447,6 +506,13 @@ def _proxy(method, *args, **kwargs):
             poll_interval=poll_interval,
             disabled=disabled,
             disable_geoip=disable_geoip,
+            feature_flags_request_timeout_seconds=feature_flags_request_timeout_seconds,
+            super_properties=super_properties,
+            # TODO: Currently this monitoring begins only when the Client is initialised (which happens when you do something with the SDK)
+            # This kind of initialisation is very annoying for exception capture. We need to figure out a way around this,
+            # or deprecate this proxy option fully (it's already in the process of deprecation, no new clients should be using this method since like 5-6 months)
+            enable_exception_autocapture=enable_exception_autocapture,
+            exception_autocapture_integrations=exception_autocapture_integrations,
         )
 
     # always set incase user changes it

posthog/ai/anthropic/__init__.py

@@ -0,0 +1,12 @@
+from .anthropic import Anthropic
+from .anthropic_async import AsyncAnthropic
+from .anthropic_providers import AnthropicBedrock, AnthropicVertex, AsyncAnthropicBedrock, AsyncAnthropicVertex
+
+__all__ = [
+    "Anthropic",
+    "AsyncAnthropic",
+    "AnthropicBedrock",
+    "AsyncAnthropicBedrock",
+    "AnthropicVertex",
+    "AsyncAnthropicVertex",
+]

posthog/ai/anthropic/anthropic.py

@@ -0,0 +1,202 @@
+try:
+    import anthropic
+    from anthropic.resources import Messages
+except ImportError:
+    raise ModuleNotFoundError("Please install the Anthropic SDK to use this feature: 'pip install anthropic'")
+
+import time
+import uuid
+from typing import Any, Dict, Optional
+
+from posthog.ai.utils import call_llm_and_track_usage, get_model_params, merge_system_prompt, with_privacy_mode
+from posthog.client import Client as PostHogClient
+
+
+class Anthropic(anthropic.Anthropic):
+    """
+    A wrapper around the Anthropic SDK that automatically sends LLM usage events to PostHog.
+    """
+
+    _ph_client: PostHogClient
+
+    def __init__(self, posthog_client: PostHogClient, **kwargs):
+        """
+        Args:
+            posthog_client: PostHog client for tracking usage
+            **kwargs: Additional arguments passed to the Anthropic client
+        """
+        super().__init__(**kwargs)
+        self._ph_client = posthog_client
+        self.messages = WrappedMessages(self)
+
+
+class WrappedMessages(Messages):
+    _client: Anthropic
+
+    def create(
+        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,
+    ):
+        """
+        Create a message using Anthropic's API while tracking 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 redact sensitive information in tracking
+            posthog_groups: Optional group analytics properties
+            **kwargs: Arguments passed to Anthropic's messages.create
+        """
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        if kwargs.get("stream", False):
+            return self._create_streaming(
+                posthog_distinct_id,
+                posthog_trace_id,
+                posthog_properties,
+                posthog_privacy_mode,
+                posthog_groups,
+                **kwargs,
+            )
+
+        return call_llm_and_track_usage(
+            posthog_distinct_id,
+            self._client._ph_client,
+            "anthropic",
+            posthog_trace_id,
+            posthog_properties,
+            posthog_privacy_mode,
+            posthog_groups,
+            self._client.base_url,
+            super().create,
+            **kwargs,
+        )
+
+    def stream(
+        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,
+    ):
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        return self._create_streaming(
+            posthog_distinct_id,
+            posthog_trace_id,
+            posthog_properties,
+            posthog_privacy_mode,
+            posthog_groups,
+            **kwargs,
+        )
+
+    def _create_streaming(
+        self,
+        posthog_distinct_id: Optional[str],
+        posthog_trace_id: Optional[str],
+        posthog_properties: Optional[Dict[str, Any]],
+        posthog_privacy_mode: bool,
+        posthog_groups: Optional[Dict[str, Any]],
+        **kwargs: Any,
+    ):
+        start_time = time.time()
+        usage_stats: Dict[str, int] = {"input_tokens": 0, "output_tokens": 0}
+        accumulated_content = []
+        response = super().create(**kwargs)
+
+        def generator():
+            nonlocal usage_stats
+            nonlocal accumulated_content
+            try:
+                for event in response:
+                    if hasattr(event, "usage") and event.usage:
+                        usage_stats = {
+                            k: getattr(event.usage, k, 0)
+                            for k in [
+                                "input_tokens",
+                                "output_tokens",
+                            ]
+                        }
+
+                    if hasattr(event, "content") and event.content:
+                        accumulated_content.append(event.content)
+
+                    yield event
+
+            finally:
+                end_time = time.time()
+                latency = end_time - start_time
+                output = "".join(accumulated_content)
+
+                self._capture_streaming_event(
+                    posthog_distinct_id,
+                    posthog_trace_id,
+                    posthog_properties,
+                    posthog_privacy_mode,
+                    posthog_groups,
+                    kwargs,
+                    usage_stats,
+                    latency,
+                    output,
+                )
+
+        return generator()
+
+    def _capture_streaming_event(
+        self,
+        posthog_distinct_id: Optional[str],
+        posthog_trace_id: Optional[str],
+        posthog_properties: Optional[Dict[str, Any]],
+        posthog_privacy_mode: bool,
+        posthog_groups: Optional[Dict[str, Any]],
+        kwargs: Dict[str, Any],
+        usage_stats: Dict[str, int],
+        latency: float,
+        output: str,
+    ):
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        event_properties = {
+            "$ai_provider": "anthropic",
+            "$ai_model": kwargs.get("model"),
+            "$ai_model_parameters": get_model_params(kwargs),
+            "$ai_input": with_privacy_mode(
+                self._client._ph_client,
+                posthog_privacy_mode,
+                merge_system_prompt(kwargs, "anthropic"),
+            ),
+            "$ai_output_choices": with_privacy_mode(
+                self._client._ph_client,
+                posthog_privacy_mode,
+                [{"content": output, "role": "assistant"}],
+            ),
+            "$ai_http_status": 200,
+            "$ai_input_tokens": usage_stats.get("input_tokens", 0),
+            "$ai_output_tokens": usage_stats.get("output_tokens", 0),
+            "$ai_latency": latency,
+            "$ai_trace_id": posthog_trace_id,
+            "$ai_base_url": str(self._client.base_url),
+            **(posthog_properties or {}),
+        }
+
+        if posthog_distinct_id is None:
+            event_properties["$process_person_profile"] = False
+
+        if hasattr(self._client._ph_client, "capture"):
+            self._client._ph_client.capture(
+                distinct_id=posthog_distinct_id or posthog_trace_id,
+                event="$ai_generation",
+                properties=event_properties,
+                groups=posthog_groups,
+            )

posthog/ai/anthropic/anthropic_async.py

@@ -0,0 +1,202 @@
+try:
+    import anthropic
+    from anthropic.resources import AsyncMessages
+except ImportError:
+    raise ModuleNotFoundError("Please install the Anthropic SDK to use this feature: 'pip install anthropic'")
+
+import time
+import uuid
+from typing import Any, Dict, Optional
+
+from posthog.ai.utils import call_llm_and_track_usage_async, get_model_params, merge_system_prompt, with_privacy_mode
+from posthog.client import Client as PostHogClient
+
+
+class AsyncAnthropic(anthropic.AsyncAnthropic):
+    """
+    An async wrapper around the Anthropic SDK that automatically sends LLM usage events to PostHog.
+    """
+
+    _ph_client: PostHogClient
+
+    def __init__(self, posthog_client: PostHogClient, **kwargs):
+        """
+        Args:
+            posthog_client: PostHog client for tracking usage
+            **kwargs: Additional arguments passed to the Anthropic client
+        """
+        super().__init__(**kwargs)
+        self._ph_client = posthog_client
+        self.messages = AsyncWrappedMessages(self)
+
+
+class AsyncWrappedMessages(AsyncMessages):
+    _client: AsyncAnthropic
+
+    async def create(
+        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,
+    ):
+        """
+        Create a message using Anthropic's API while tracking 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 redact sensitive information in tracking
+            posthog_groups: Optional group analytics properties
+            **kwargs: Arguments passed to Anthropic's messages.create
+        """
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        if kwargs.get("stream", False):
+            return await self._create_streaming(
+                posthog_distinct_id,
+                posthog_trace_id,
+                posthog_properties,
+                posthog_privacy_mode,
+                posthog_groups,
+                **kwargs,
+            )
+
+        return await call_llm_and_track_usage_async(
+            posthog_distinct_id,
+            self._client._ph_client,
+            "anthropic",
+            posthog_trace_id,
+            posthog_properties,
+            posthog_privacy_mode,
+            posthog_groups,
+            self._client.base_url,
+            super().create,
+            **kwargs,
+        )
+
+    async def stream(
+        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,
+    ):
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        return await self._create_streaming(
+            posthog_distinct_id,
+            posthog_trace_id,
+            posthog_properties,
+            posthog_privacy_mode,
+            posthog_groups,
+            **kwargs,
+        )
+
+    async def _create_streaming(
+        self,
+        posthog_distinct_id: Optional[str],
+        posthog_trace_id: Optional[str],
+        posthog_properties: Optional[Dict[str, Any]],
+        posthog_privacy_mode: bool,
+        posthog_groups: Optional[Dict[str, Any]],
+        **kwargs: Any,
+    ):
+        start_time = time.time()
+        usage_stats: Dict[str, int] = {"input_tokens": 0, "output_tokens": 0}
+        accumulated_content = []
+        response = await super().create(**kwargs)
+
+        async def generator():
+            nonlocal usage_stats
+            nonlocal accumulated_content
+            try:
+                async for event in response:
+                    if hasattr(event, "usage") and event.usage:
+                        usage_stats = {
+                            k: getattr(event.usage, k, 0)
+                            for k in [
+                                "input_tokens",
+                                "output_tokens",
+                            ]
+                        }
+
+                    if hasattr(event, "content") and event.content:
+                        accumulated_content.append(event.content)
+
+                    yield event
+
+            finally:
+                end_time = time.time()
+                latency = end_time - start_time
+                output = "".join(accumulated_content)
+
+                await self._capture_streaming_event(
+                    posthog_distinct_id,
+                    posthog_trace_id,
+                    posthog_properties,
+                    posthog_privacy_mode,
+                    posthog_groups,
+                    kwargs,
+                    usage_stats,
+                    latency,
+                    output,
+                )
+
+        return generator()
+
+    async def _capture_streaming_event(
+        self,
+        posthog_distinct_id: Optional[str],
+        posthog_trace_id: Optional[str],
+        posthog_properties: Optional[Dict[str, Any]],
+        posthog_privacy_mode: bool,
+        posthog_groups: Optional[Dict[str, Any]],
+        kwargs: Dict[str, Any],
+        usage_stats: Dict[str, int],
+        latency: float,
+        output: str,
+    ):
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        event_properties = {
+            "$ai_provider": "anthropic",
+            "$ai_model": kwargs.get("model"),
+            "$ai_model_parameters": get_model_params(kwargs),
+            "$ai_input": with_privacy_mode(
+                self._client._ph_client,
+                posthog_privacy_mode,
+                merge_system_prompt(kwargs, "anthropic"),
+            ),
+            "$ai_output_choices": with_privacy_mode(
+                self._client._ph_client,
+                posthog_privacy_mode,
+                [{"content": output, "role": "assistant"}],
+            ),
+            "$ai_http_status": 200,
+            "$ai_input_tokens": usage_stats.get("input_tokens", 0),
+            "$ai_output_tokens": usage_stats.get("output_tokens", 0),
+            "$ai_latency": latency,
+            "$ai_trace_id": posthog_trace_id,
+            "$ai_base_url": str(self._client.base_url),
+            **(posthog_properties or {}),
+        }
+
+        if posthog_distinct_id is None:
+            event_properties["$process_person_profile"] = False
+
+        if hasattr(self._client._ph_client, "capture"):
+            self._client._ph_client.capture(
+                distinct_id=posthog_distinct_id or posthog_trace_id,
+                event="$ai_generation",
+                properties=event_properties,
+                groups=posthog_groups,
+            )

posthog/ai/anthropic/anthropic_providers.py

@@ -0,0 +1,60 @@
+try:
+    import anthropic
+except ImportError:
+    raise ModuleNotFoundError("Please install the Anthropic SDK to use this feature: 'pip install anthropic'")
+
+from posthog.ai.anthropic.anthropic import WrappedMessages
+from posthog.ai.anthropic.anthropic_async import AsyncWrappedMessages
+from posthog.client import Client as PostHogClient
+
+
+class AnthropicBedrock(anthropic.AnthropicBedrock):
+    """
+    A wrapper around the Anthropic Bedrock SDK that automatically sends LLM usage events to PostHog.
+    """
+
+    _ph_client: PostHogClient
+
+    def __init__(self, posthog_client: PostHogClient, **kwargs):
+        super().__init__(**kwargs)
+        self._ph_client = posthog_client
+        self.messages = WrappedMessages(self)
+
+
+class AsyncAnthropicBedrock(anthropic.AsyncAnthropicBedrock):
+    """
+    A wrapper around the Anthropic Bedrock SDK that automatically sends LLM usage events to PostHog.
+    """
+
+    _ph_client: PostHogClient
+
+    def __init__(self, posthog_client: PostHogClient, **kwargs):
+        super().__init__(**kwargs)
+        self._ph_client = posthog_client
+        self.messages = AsyncWrappedMessages(self)
+
+
+class AnthropicVertex(anthropic.AnthropicVertex):
+    """
+    A wrapper around the Anthropic Vertex SDK that automatically sends LLM usage events to PostHog.
+    """
+
+    _ph_client: PostHogClient
+
+    def __init__(self, posthog_client: PostHogClient, **kwargs):
+        super().__init__(**kwargs)
+        self._ph_client = posthog_client
+        self.messages = WrappedMessages(self)
+
+
+class AsyncAnthropicVertex(anthropic.AsyncAnthropicVertex):
+    """
+    A wrapper around the Anthropic Vertex SDK that automatically sends LLM usage events to PostHog.
+    """
+
+    _ph_client: PostHogClient
+
+    def __init__(self, posthog_client: PostHogClient, **kwargs):
+        super().__init__(**kwargs)
+        self._ph_client = posthog_client
+        self.messages = AsyncWrappedMessages(self)

posthog/ai/langchain/__init__.py

@@ -0,0 +1,3 @@
+from .callbacks import CallbackHandler
+
+__all__ = ["CallbackHandler"]

posthog/ai/langchain/callbacks.py

@@ -0,0 +1,419 @@
+try:
+    import langchain  # noqa: F401
+except ImportError:
+    raise ModuleNotFoundError("Please install LangChain to use this feature: 'pip install langchain'")
+
+import logging
+import time
+import uuid
+from typing import (
+    Any,
+    Dict,
+    List,
+    Optional,
+    Tuple,
+    TypedDict,
+    Union,
+    cast,
+)
+from uuid import UUID
+
+from langchain.callbacks.base import BaseCallbackHandler
+from langchain_core.messages import AIMessage, BaseMessage, FunctionMessage, HumanMessage, SystemMessage, ToolMessage
+from langchain_core.outputs import ChatGeneration, LLMResult
+from pydantic import BaseModel
+
+from posthog.ai.utils import get_model_params, with_privacy_mode
+from posthog.client import Client
+
+log = logging.getLogger("posthog")
+
+
+class RunMetadata(TypedDict, total=False):
+    messages: Union[List[Dict[str, Any]], List[str]]
+    provider: str
+    model: str
+    model_params: Dict[str, Any]
+    base_url: str
+    start_time: float
+    end_time: float
+
+
+RunStorage = Dict[UUID, RunMetadata]
+
+
+class CallbackHandler(BaseCallbackHandler):
+    """
+    A callback handler for LangChain that sends events to PostHog LLM Observability.
+    """
+
+    _client: Client
+    """PostHog client instance."""
+    _distinct_id: Optional[Union[str, int, float, UUID]]
+    """Distinct ID of the user to associate the trace with."""
+    _trace_id: Optional[Union[str, int, float, UUID]]
+    """Global trace ID to be sent with every event. Otherwise, the top-level run ID is used."""
+    _properties: Optional[Dict[str, Any]]
+    """Global properties to be sent with every event."""
+    _runs: RunStorage
+    """Mapping of run IDs to run metadata as run metadata is only available on the start of generation."""
+    _parent_tree: Dict[UUID, UUID]
+    """
+    A dictionary that maps chain run IDs to their parent chain run IDs (parent pointer tree),
+    so the top level can be found from a bottom-level run ID.
+    """
+
+    def __init__(
+        self,
+        client: Client,
+        distinct_id: Optional[Union[str, int, float, UUID]] = None,
+        trace_id: Optional[Union[str, int, float, UUID]] = None,
+        properties: Optional[Dict[str, Any]] = None,
+        privacy_mode: bool = False,
+        groups: Optional[Dict[str, Any]] = None,
+    ):
+        """
+        Args:
+            client: PostHog client instance.
+            distinct_id: Optional distinct ID of the user to associate the trace with.
+            trace_id: Optional trace ID to use for the event.
+            properties: Optional additional metadata to use for the trace.
+            privacy_mode: Whether to redact the input and output of the trace.
+            groups: Optional additional PostHog groups to use for the trace.
+        """
+        self._client = client
+        self._distinct_id = distinct_id
+        self._trace_id = trace_id
+        self._properties = properties or {}
+        self._privacy_mode = privacy_mode
+        self._groups = groups or {}
+        self._runs = {}
+        self._parent_tree = {}
+
+    def on_chain_start(
+        self,
+        serialized: Dict[str, Any],
+        inputs: Dict[str, Any],
+        *,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        **kwargs,
+    ):
+        self._set_parent_of_run(run_id, parent_run_id)
+
+    def on_chat_model_start(
+        self,
+        serialized: Dict[str, Any],
+        messages: List[List[BaseMessage]],
+        *,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        **kwargs,
+    ):
+        self._set_parent_of_run(run_id, parent_run_id)
+        input = [_convert_message_to_dict(message) for row in messages for message in row]
+        self._set_run_metadata(serialized, run_id, input, **kwargs)
+
+    def on_llm_start(
+        self,
+        serialized: Dict[str, Any],
+        prompts: List[str],
+        *,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        **kwargs: Any,
+    ):
+        self._set_parent_of_run(run_id, parent_run_id)
+        self._set_run_metadata(serialized, run_id, prompts, **kwargs)
+
+    def on_chain_end(
+        self,
+        outputs: Dict[str, Any],
+        *,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        tags: Optional[List[str]] = None,
+        **kwargs: Any,
+    ):
+        self._pop_parent_of_run(run_id)
+
+    def on_llm_end(
+        self,
+        response: LLMResult,
+        *,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        tags: Optional[List[str]] = None,
+        **kwargs: Any,
+    ):
+        """
+        The callback works for both streaming and non-streaming runs. For streaming runs, the chain must set `stream_usage=True` in the LLM.
+        """
+        trace_id = self._get_trace_id(run_id)
+        self._pop_parent_of_run(run_id)
+        run = self._pop_run_metadata(run_id)
+        if not run:
+            return
+
+        latency = run.get("end_time", 0) - run.get("start_time", 0)
+        input_tokens, output_tokens = _parse_usage(response)
+
+        generation_result = response.generations[-1]
+        if isinstance(generation_result[-1], ChatGeneration):
+            output = [
+                _convert_message_to_dict(cast(ChatGeneration, generation).message) for generation in generation_result
+            ]
+        else:
+            output = [_extract_raw_esponse(generation) for generation in generation_result]
+
+        event_properties = {
+            "$ai_provider": run.get("provider"),
+            "$ai_model": run.get("model"),
+            "$ai_model_parameters": run.get("model_params"),
+            "$ai_input": with_privacy_mode(self._client, self._privacy_mode, run.get("messages")),
+            "$ai_output_choices": with_privacy_mode(self._client, self._privacy_mode, output),
+            "$ai_http_status": 200,
+            "$ai_input_tokens": input_tokens,
+            "$ai_output_tokens": output_tokens,
+            "$ai_latency": latency,
+            "$ai_trace_id": trace_id,
+            "$ai_base_url": run.get("base_url"),
+            **self._properties,
+        }
+        if self._distinct_id is None:
+            event_properties["$process_person_profile"] = False
+        self._client.capture(
+            distinct_id=self._distinct_id or trace_id,
+            event="$ai_generation",
+            properties=event_properties,
+            groups=self._groups,
+        )
+
+    def on_chain_error(
+        self,
+        error: BaseException,
+        *,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        **kwargs: Any,
+    ):
+        self._pop_parent_of_run(run_id)
+
+    def on_llm_error(
+        self,
+        error: BaseException,
+        *,
+        run_id: UUID,
+        parent_run_id: Optional[UUID] = None,
+        tags: Optional[List[str]] = None,
+        **kwargs: Any,
+    ):
+        trace_id = self._get_trace_id(run_id)
+        self._pop_parent_of_run(run_id)
+        run = self._pop_run_metadata(run_id)
+        if not run:
+            return
+
+        latency = run.get("end_time", 0) - run.get("start_time", 0)
+        event_properties = {
+            "$ai_provider": run.get("provider"),
+            "$ai_model": run.get("model"),
+            "$ai_model_parameters": run.get("model_params"),
+            "$ai_input": with_privacy_mode(self._client, self._privacy_mode, run.get("messages")),
+            "$ai_http_status": _get_http_status(error),
+            "$ai_latency": latency,
+            "$ai_trace_id": trace_id,
+            "$ai_base_url": run.get("base_url"),
+            **self._properties,
+        }
+        if self._distinct_id is None:
+            event_properties["$process_person_profile"] = False
+        self._client.capture(
+            distinct_id=self._distinct_id or trace_id,
+            event="$ai_generation",
+            properties=event_properties,
+            groups=self._groups,
+        )
+
+    def _set_parent_of_run(self, run_id: UUID, parent_run_id: Optional[UUID] = None):
+        """
+        Set the parent run ID for a chain run. If there is no parent, the run is the root.
+        """
+        if parent_run_id is not None:
+            self._parent_tree[run_id] = parent_run_id
+
+    def _pop_parent_of_run(self, run_id: UUID):
+        """
+        Remove the parent run ID for a chain run.
+        """
+        try:
+            self._parent_tree.pop(run_id)
+        except KeyError:
+            pass
+
+    def _find_root_run(self, run_id: UUID) -> UUID:
+        """
+        Finds the root ID of a chain run.
+        """
+        id: UUID = run_id
+        while id in self._parent_tree:
+            id = self._parent_tree[id]
+        return id
+
+    def _set_run_metadata(
+        self,
+        serialized: Dict[str, Any],
+        run_id: UUID,
+        messages: Union[List[Dict[str, Any]], List[str]],
+        metadata: Optional[Dict[str, Any]] = None,
+        invocation_params: Optional[Dict[str, Any]] = None,
+        **kwargs,
+    ):
+        run: RunMetadata = {
+            "messages": messages,
+            "start_time": time.time(),
+        }
+        if isinstance(invocation_params, dict):
+            run["model_params"] = get_model_params(invocation_params)
+        if isinstance(metadata, dict):
+            if model := metadata.get("ls_model_name"):
+                run["model"] = model
+            if provider := metadata.get("ls_provider"):
+                run["provider"] = provider
+        try:
+            base_url = serialized["kwargs"]["openai_api_base"]
+            if base_url is not None:
+                run["base_url"] = base_url
+        except KeyError:
+            pass
+        self._runs[run_id] = run
+
+    def _pop_run_metadata(self, run_id: UUID) -> Optional[RunMetadata]:
+        end_time = time.time()
+        try:
+            run = self._runs.pop(run_id)
+        except KeyError:
+            log.warning(f"No run metadata found for run {run_id}")
+            return None
+        run["end_time"] = end_time
+        return run
+
+    def _get_trace_id(self, run_id: UUID):
+        trace_id = self._trace_id or self._find_root_run(run_id)
+        if not trace_id:
+            trace_id = uuid.uuid4()
+        return trace_id
+
+
+def _extract_raw_esponse(last_response):
+    """Extract the response from the last response of the LLM call."""
+    # We return the text of the response if not empty
+    if last_response.text is not None and last_response.text.strip() != "":
+        return last_response.text.strip()
+    elif hasattr(last_response, "message"):
+        # Additional kwargs contains the response in case of tool usage
+        return last_response.message.additional_kwargs
+    else:
+        # Not tool usage, some LLM responses can be simply empty
+        return ""
+
+
+def _convert_message_to_dict(message: BaseMessage) -> Dict[str, Any]:
+    # assistant message
+    if isinstance(message, HumanMessage):
+        message_dict = {"role": "user", "content": message.content}
+    elif isinstance(message, AIMessage):
+        message_dict = {"role": "assistant", "content": message.content}
+    elif isinstance(message, SystemMessage):
+        message_dict = {"role": "system", "content": message.content}
+    elif isinstance(message, ToolMessage):
+        message_dict = {"role": "tool", "content": message.content}
+    elif isinstance(message, FunctionMessage):
+        message_dict = {"role": "function", "content": message.content}
+    else:
+        message_dict = {"role": message.type, "content": str(message.content)}
+
+    if message.additional_kwargs:
+        message_dict.update(message.additional_kwargs)
+
+    return message_dict
+
+
+def _parse_usage_model(usage: Union[BaseModel, Dict]) -> Tuple[Union[int, None], Union[int, None]]:
+    if isinstance(usage, BaseModel):
+        usage = usage.__dict__
+
+    conversion_list = [
+        # https://pypi.org/project/langchain-anthropic/ (works also for Bedrock-Anthropic)
+        ("input_tokens", "input"),
+        ("output_tokens", "output"),
+        # https://cloud.google.com/vertex-ai/generative-ai/docs/multimodal/get-token-count
+        ("prompt_token_count", "input"),
+        ("candidates_token_count", "output"),
+        # Bedrock: https://docs.aws.amazon.com/bedrock/latest/userguide/monitoring-cw.html#runtime-cloudwatch-metrics
+        ("inputTokenCount", "input"),
+        ("outputTokenCount", "output"),
+        # langchain-ibm https://pypi.org/project/langchain-ibm/
+        ("input_token_count", "input"),
+        ("generated_token_count", "output"),
+    ]
+
+    parsed_usage = {}
+    for model_key, type_key in conversion_list:
+        if model_key in usage:
+            captured_count = usage[model_key]
+            final_count = (
+                sum(captured_count) if isinstance(captured_count, list) else captured_count
+            )  # For Bedrock, the token count is a list when streamed
+
+            parsed_usage[type_key] = final_count
+
+    return parsed_usage.get("input"), parsed_usage.get("output")
+
+
+def _parse_usage(response: LLMResult):
+    # langchain-anthropic uses the usage field
+    llm_usage_keys = ["token_usage", "usage"]
+    llm_usage: Tuple[Union[int, None], Union[int, None]] = (None, None)
+    if response.llm_output is not None:
+        for key in llm_usage_keys:
+            if response.llm_output.get(key):
+                llm_usage = _parse_usage_model(response.llm_output[key])
+                break
+
+    if hasattr(response, "generations"):
+        for generation in response.generations:
+            for generation_chunk in generation:
+                if generation_chunk.generation_info and ("usage_metadata" in generation_chunk.generation_info):
+                    llm_usage = _parse_usage_model(generation_chunk.generation_info["usage_metadata"])
+                    break
+
+                message_chunk = getattr(generation_chunk, "message", {})
+                response_metadata = getattr(message_chunk, "response_metadata", {})
+
+                bedrock_anthropic_usage = (
+                    response_metadata.get("usage", None)  # for Bedrock-Anthropic
+                    if isinstance(response_metadata, dict)
+                    else None
+                )
+                bedrock_titan_usage = (
+                    response_metadata.get("amazon-bedrock-invocationMetrics", None)  # for Bedrock-Titan
+                    if isinstance(response_metadata, dict)
+                    else None
+                )
+                ollama_usage = getattr(message_chunk, "usage_metadata", None)  # for Ollama
+
+                chunk_usage = bedrock_anthropic_usage or bedrock_titan_usage or ollama_usage
+                if chunk_usage:
+                    llm_usage = _parse_usage_model(chunk_usage)
+                    break
+
+    return llm_usage
+
+
+def _get_http_status(error: BaseException) -> int:
+    # OpenAI: https://github.com/openai/openai-python/blob/main/src/openai/_exceptions.py
+    # Anthropic: https://github.com/anthropics/anthropic-sdk-python/blob/main/src/anthropic/_exceptions.py
+    # Google: https://github.com/googleapis/python-api-core/blob/main/google/api_core/exceptions.py
+    status_code = getattr(error, "status_code", getattr(error, "code", 0))
+    return status_code

posthog/ai/openai/__init__.py

@@ -0,0 +1,4 @@
+from .openai import OpenAI
+from .openai_async import AsyncOpenAI
+
+__all__ = ["OpenAI", "AsyncOpenAI"]

posthog/ai/openai/openai.py

@@ -0,0 +1,251 @@
+import time
+import uuid
+from typing import Any, Dict, Optional
+
+try:
+    import openai
+    import openai.resources
+except ImportError:
+    raise ModuleNotFoundError("Please install the OpenAI SDK to use this feature: 'pip install openai'")
+
+from posthog.ai.utils import call_llm_and_track_usage, get_model_params, with_privacy_mode
+from posthog.client import Client as PostHogClient
+
+
+class OpenAI(openai.OpenAI):
+    """
+    A wrapper around the OpenAI SDK that automatically sends LLM usage events to PostHog.
+    """
+
+    _ph_client: PostHogClient
+
+    def __init__(self, posthog_client: PostHogClient, **kwargs):
+        """
+        Args:
+            api_key: OpenAI API key.
+            posthog_client: If provided, events will be captured via this client instead
+                            of the global posthog.
+            **openai_config: Any additional keyword args to set on openai (e.g. organization="xxx").
+        """
+        super().__init__(**kwargs)
+        self._ph_client = posthog_client
+        self.chat = WrappedChat(self)
+        self.embeddings = WrappedEmbeddings(self)
+
+
+class WrappedChat(openai.resources.chat.Chat):
+    _client: OpenAI
+
+    @property
+    def completions(self):
+        return WrappedCompletions(self._client)
+
+
+class WrappedCompletions(openai.resources.chat.completions.Completions):
+    _client: OpenAI
+
+    def create(
+        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,
+    ):
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        if kwargs.get("stream", False):
+            return self._create_streaming(
+                posthog_distinct_id,
+                posthog_trace_id,
+                posthog_properties,
+                posthog_privacy_mode,
+                posthog_groups,
+                **kwargs,
+            )
+
+        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,
+            super().create,
+            **kwargs,
+        )
+
+    def _create_streaming(
+        self,
+        posthog_distinct_id: Optional[str],
+        posthog_trace_id: Optional[str],
+        posthog_properties: Optional[Dict[str, Any]],
+        posthog_privacy_mode: bool,
+        posthog_groups: Optional[Dict[str, Any]],
+        **kwargs: Any,
+    ):
+        start_time = time.time()
+        usage_stats: Dict[str, int] = {}
+        accumulated_content = []
+        if "stream_options" not in kwargs:
+            kwargs["stream_options"] = {}
+        kwargs["stream_options"]["include_usage"] = True
+        response = super().create(**kwargs)
+
+        def generator():
+            nonlocal usage_stats
+            nonlocal accumulated_content
+            try:
+                for chunk in response:
+                    if hasattr(chunk, "usage") and chunk.usage:
+                        usage_stats = {
+                            k: getattr(chunk.usage, k, 0)
+                            for k in [
+                                "prompt_tokens",
+                                "completion_tokens",
+                                "total_tokens",
+                            ]
+                        }
+
+                    if hasattr(chunk, "choices") and chunk.choices and len(chunk.choices) > 0:
+                        content = chunk.choices[0].delta.content
+                        if content:
+                            accumulated_content.append(content)
+
+                    yield chunk
+
+            finally:
+                end_time = time.time()
+                latency = end_time - start_time
+                output = "".join(accumulated_content)
+                self._capture_streaming_event(
+                    posthog_distinct_id,
+                    posthog_trace_id,
+                    posthog_properties,
+                    posthog_privacy_mode,
+                    posthog_groups,
+                    kwargs,
+                    usage_stats,
+                    latency,
+                    output,
+                )
+
+        return generator()
+
+    def _capture_streaming_event(
+        self,
+        posthog_distinct_id: Optional[str],
+        posthog_trace_id: Optional[str],
+        posthog_properties: Optional[Dict[str, Any]],
+        posthog_privacy_mode: bool,
+        posthog_groups: Optional[Dict[str, Any]],
+        kwargs: Dict[str, Any],
+        usage_stats: Dict[str, int],
+        latency: float,
+        output: str,
+    ):
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        event_properties = {
+            "$ai_provider": "openai",
+            "$ai_model": kwargs.get("model"),
+            "$ai_model_parameters": get_model_params(kwargs),
+            "$ai_input": with_privacy_mode(self._client._ph_client, posthog_privacy_mode, kwargs.get("messages")),
+            "$ai_output_choices": with_privacy_mode(
+                self._client._ph_client,
+                posthog_privacy_mode,
+                [{"content": output, "role": "assistant"}],
+            ),
+            "$ai_http_status": 200,
+            "$ai_input_tokens": usage_stats.get("prompt_tokens", 0),
+            "$ai_output_tokens": usage_stats.get("completion_tokens", 0),
+            "$ai_latency": latency,
+            "$ai_trace_id": posthog_trace_id,
+            "$ai_base_url": str(self._client.base_url),
+            **posthog_properties,
+        }
+
+        if posthog_distinct_id is None:
+            event_properties["$process_person_profile"] = False
+
+        if hasattr(self._client._ph_client, "capture"):
+            self._client._ph_client.capture(
+                distinct_id=posthog_distinct_id or posthog_trace_id,
+                event="$ai_generation",
+                properties=event_properties,
+                groups=posthog_groups,
+            )
+
+
+class WrappedEmbeddings(openai.resources.embeddings.Embeddings):
+    _client: OpenAI
+
+    def create(
+        self,
+        posthog_distinct_id: Optional[str] = None,
+        posthog_trace_id: Optional[str] = None,
+        posthog_properties: Optional[Dict[str, Any]] = None,
+        posthog_privacy_mode: bool = False,
+        posthog_groups: Optional[Dict[str, Any]] = None,
+        **kwargs: Any,
+    ):
+        """
+        Create an embedding using OpenAI's 'embeddings.create' method, but also track usage in PostHog.
+
+        Args:
+            posthog_distinct_id: Optional ID to associate with the usage event.
+            posthog_trace_id: Optional trace UUID for linking events.
+            posthog_properties: Optional dictionary of extra properties to include in the event.
+            **kwargs: Any additional parameters for the OpenAI Embeddings API.
+
+        Returns:
+            The response from OpenAI's embeddings.create call.
+        """
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        start_time = time.time()
+        response = super().create(**kwargs)
+        end_time = time.time()
+
+        # Extract usage statistics if available
+        usage_stats = {}
+        if hasattr(response, "usage") and response.usage:
+            usage_stats = {
+                "prompt_tokens": getattr(response.usage, "prompt_tokens", 0),
+                "total_tokens": getattr(response.usage, "total_tokens", 0),
+            }
+
+        latency = end_time - start_time
+
+        # Build the event properties
+        event_properties = {
+            "$ai_provider": "openai",
+            "$ai_model": kwargs.get("model"),
+            "$ai_input": with_privacy_mode(self._client._ph_client, posthog_privacy_mode, kwargs.get("input")),
+            "$ai_http_status": 200,
+            "$ai_input_tokens": usage_stats.get("prompt_tokens", 0),
+            "$ai_latency": latency,
+            "$ai_trace_id": posthog_trace_id,
+            "$ai_base_url": str(self._client.base_url),
+            **posthog_properties,
+        }
+
+        if posthog_distinct_id is None:
+            event_properties["$process_person_profile"] = False
+
+        # Send capture event for embeddings
+        if hasattr(self._client._ph_client, "capture"):
+            self._client._ph_client.capture(
+                distinct_id=posthog_distinct_id or posthog_trace_id,
+                event="$ai_embedding",
+                properties=event_properties,
+                groups=posthog_groups,
+            )
+
+        return response

posthog/ai/openai/openai_async.py

@@ -0,0 +1,250 @@
+import time
+import uuid
+from typing import Any, Dict, Optional
+
+try:
+    import openai
+    import openai.resources
+except ImportError:
+    raise ModuleNotFoundError("Please install the OpenAI SDK to use this feature: 'pip install openai'")
+
+from posthog.ai.utils import call_llm_and_track_usage_async, get_model_params, with_privacy_mode
+from posthog.client import Client as PostHogClient
+
+
+class AsyncOpenAI(openai.AsyncOpenAI):
+    """
+    An async wrapper around the OpenAI SDK that automatically sends LLM usage events to PostHog.
+    """
+
+    _ph_client: PostHogClient
+
+    def __init__(self, posthog_client: PostHogClient, **kwargs):
+        """
+        Args:
+            api_key: OpenAI API key.
+            posthog_client: If provided, events will be captured via this client instance.
+            **openai_config: Additional keyword args (e.g. organization="xxx").
+        """
+        super().__init__(**kwargs)
+        self._ph_client = posthog_client
+        self.chat = WrappedChat(self)
+        self.embeddings = WrappedEmbeddings(self)
+
+
+class WrappedChat(openai.resources.chat.AsyncChat):
+    _client: AsyncOpenAI
+
+    @property
+    def completions(self):
+        return WrappedCompletions(self._client)
+
+
+class WrappedCompletions(openai.resources.chat.completions.AsyncCompletions):
+    _client: AsyncOpenAI
+
+    async def create(
+        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,
+    ):
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        # If streaming, handle streaming specifically
+        if kwargs.get("stream", False):
+            return await self._create_streaming(
+                posthog_distinct_id,
+                posthog_trace_id,
+                posthog_properties,
+                posthog_privacy_mode,
+                posthog_groups,
+                **kwargs,
+            )
+
+        response = await call_llm_and_track_usage_async(
+            posthog_distinct_id,
+            self._client._ph_client,
+            "openai",
+            posthog_trace_id,
+            posthog_properties,
+            self._client.base_url,
+            super().create,
+            **kwargs,
+        )
+        return response
+
+    async def _create_streaming(
+        self,
+        posthog_distinct_id: Optional[str],
+        posthog_trace_id: Optional[str],
+        posthog_properties: Optional[Dict[str, Any]],
+        posthog_privacy_mode: bool = False,
+        posthog_groups: Optional[Dict[str, Any]] = None,
+        **kwargs: Any,
+    ):
+        start_time = time.time()
+        usage_stats: Dict[str, int] = {}
+        accumulated_content = []
+        if "stream_options" not in kwargs:
+            kwargs["stream_options"] = {}
+        kwargs["stream_options"]["include_usage"] = True
+        response = await super().create(**kwargs)
+
+        async def async_generator():
+            nonlocal usage_stats, accumulated_content
+            try:
+                async for chunk in response:
+                    if hasattr(chunk, "usage") and chunk.usage:
+                        usage_stats = {
+                            k: getattr(chunk.usage, k, 0)
+                            for k in [
+                                "prompt_tokens",
+                                "completion_tokens",
+                                "total_tokens",
+                            ]
+                        }
+                    if hasattr(chunk, "choices") and chunk.choices and len(chunk.choices) > 0:
+                        content = chunk.choices[0].delta.content
+                        if content:
+                            accumulated_content.append(content)
+
+                    yield chunk
+
+            finally:
+                end_time = time.time()
+                latency = end_time - start_time
+                output = "".join(accumulated_content)
+                await self._capture_streaming_event(
+                    posthog_distinct_id,
+                    posthog_trace_id,
+                    posthog_properties,
+                    posthog_privacy_mode,
+                    posthog_groups,
+                    kwargs,
+                    usage_stats,
+                    latency,
+                    output,
+                )
+
+        return async_generator()
+
+    async def _capture_streaming_event(
+        self,
+        posthog_distinct_id: Optional[str],
+        posthog_trace_id: Optional[str],
+        posthog_properties: Optional[Dict[str, Any]],
+        posthog_privacy_mode: bool,
+        posthog_groups: Optional[Dict[str, Any]],
+        kwargs: Dict[str, Any],
+        usage_stats: Dict[str, int],
+        latency: float,
+        output: str,
+    ):
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        event_properties = {
+            "$ai_provider": "openai",
+            "$ai_model": kwargs.get("model"),
+            "$ai_model_parameters": get_model_params(kwargs),
+            "$ai_input": with_privacy_mode(self._client._ph_client, posthog_privacy_mode, kwargs.get("messages")),
+            "$ai_output_choices": with_privacy_mode(
+                self._client._ph_client,
+                posthog_privacy_mode,
+                [{"content": output, "role": "assistant"}],
+            ),
+            "$ai_http_status": 200,
+            "$ai_input_tokens": usage_stats.get("prompt_tokens", 0),
+            "$ai_output_tokens": usage_stats.get("completion_tokens", 0),
+            "$ai_latency": latency,
+            "$ai_trace_id": posthog_trace_id,
+            "$ai_base_url": str(self._client.base_url),
+            **posthog_properties,
+        }
+
+        if posthog_distinct_id is None:
+            event_properties["$process_person_profile"] = False
+
+        if hasattr(self._client._ph_client, "capture"):
+            self._client._ph_client.capture(
+                distinct_id=posthog_distinct_id or posthog_trace_id,
+                event="$ai_generation",
+                properties=event_properties,
+                groups=posthog_groups,
+            )
+
+
+class WrappedEmbeddings(openai.resources.embeddings.AsyncEmbeddings):
+    _client: AsyncOpenAI
+
+    async def create(
+        self,
+        posthog_distinct_id: Optional[str] = None,
+        posthog_trace_id: Optional[str] = None,
+        posthog_properties: Optional[Dict[str, Any]] = None,
+        posthog_privacy_mode: bool = False,
+        posthog_groups: Optional[Dict[str, Any]] = None,
+        **kwargs: Any,
+    ):
+        """
+        Create an embedding using OpenAI's 'embeddings.create' method, but also track usage in PostHog.
+
+        Args:
+            posthog_distinct_id: Optional ID to associate with the usage event.
+            posthog_trace_id: Optional trace UUID for linking events.
+            posthog_properties: Optional dictionary of extra properties to include in the event.
+            posthog_privacy_mode: Whether to store input and output in PostHog.
+            posthog_groups: Optional dictionary of groups to include in the event.
+            **kwargs: Any additional parameters for the OpenAI Embeddings API.
+
+        Returns:
+            The response from OpenAI's embeddings.create call.
+        """
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        start_time = time.time()
+        response = await super().create(**kwargs)
+        end_time = time.time()
+
+        # Extract usage statistics if available
+        usage_stats = {}
+        if hasattr(response, "usage") and response.usage:
+            usage_stats = {
+                "prompt_tokens": getattr(response.usage, "prompt_tokens", 0),
+                "total_tokens": getattr(response.usage, "total_tokens", 0),
+            }
+
+        latency = end_time - start_time
+
+        # Build the event properties
+        event_properties = {
+            "$ai_provider": "openai",
+            "$ai_model": kwargs.get("model"),
+            "$ai_input": with_privacy_mode(self._client._ph_client, posthog_privacy_mode, kwargs.get("input")),
+            "$ai_http_status": 200,
+            "$ai_input_tokens": usage_stats.get("prompt_tokens", 0),
+            "$ai_latency": latency,
+            "$ai_trace_id": posthog_trace_id,
+            "$ai_base_url": str(self._client.base_url),
+            **posthog_properties,
+        }
+
+        if posthog_distinct_id is None:
+            event_properties["$process_person_profile"] = False
+
+        # Send capture event for embeddings
+        if hasattr(self._client._ph_client, "capture"):
+            self._client._ph_client.capture(
+                distinct_id=posthog_distinct_id or posthog_trace_id,
+                event="$ai_embedding",
+                properties=event_properties,
+                groups=posthog_groups,
+            )
+
+        return response

posthog/ai/utils.py

@@ -0,0 +1,245 @@
+import time
+import uuid
+from typing import Any, Callable, Dict, Optional
+
+from httpx import URL
+
+from posthog.client import Client as PostHogClient
+
+
+def get_model_params(kwargs: Dict[str, Any]) -> Dict[str, Any]:
+    """
+    Extracts model parameters from the kwargs dictionary.
+    """
+    model_params = {}
+    for param in [
+        "temperature",
+        "max_tokens",  # Deprecated field
+        "max_completion_tokens",
+        "top_p",
+        "frequency_penalty",
+        "presence_penalty",
+        "n",
+        "stop",
+        "stream",  # OpenAI-specific field
+        "streaming",  # Anthropic-specific field
+    ]:
+        if param in kwargs and kwargs[param] is not None:
+            model_params[param] = kwargs[param]
+    return model_params
+
+
+def get_usage(response, provider: str) -> Dict[str, Any]:
+    if provider == "anthropic":
+        return {
+            "input_tokens": response.usage.input_tokens,
+            "output_tokens": response.usage.output_tokens,
+        }
+    elif provider == "openai":
+        return {
+            "input_tokens": response.usage.prompt_tokens,
+            "output_tokens": response.usage.completion_tokens,
+        }
+    return {
+        "input_tokens": 0,
+        "output_tokens": 0,
+    }
+
+
+def format_response(response, provider: str):
+    """
+    Format a regular (non-streaming) response.
+    """
+    output = []
+    if response is None:
+        return output
+    if provider == "anthropic":
+        return format_response_anthropic(response)
+    elif provider == "openai":
+        return format_response_openai(response)
+    return output
+
+
+def format_response_anthropic(response):
+    output = []
+    for choice in response.content:
+        if choice.text:
+            output.append(
+                {
+                    "role": "assistant",
+                    "content": choice.text,
+                }
+            )
+    return output
+
+
+def format_response_openai(response):
+    output = []
+    for choice in response.choices:
+        if choice.message.content:
+            output.append(
+                {
+                    "content": choice.message.content,
+                    "role": choice.message.role,
+                }
+            )
+    return output
+
+
+def merge_system_prompt(kwargs: Dict[str, Any], provider: str):
+    if provider != "anthropic":
+        return kwargs.get("messages")
+    messages = kwargs.get("messages") or []
+    if kwargs.get("system") is None:
+        return messages
+    return [{"role": "system", "content": kwargs.get("system")}] + messages
+
+
+def call_llm_and_track_usage(
+    posthog_distinct_id: Optional[str],
+    ph_client: PostHogClient,
+    provider: str,
+    posthog_trace_id: Optional[str],
+    posthog_properties: Optional[Dict[str, Any]],
+    posthog_privacy_mode: bool,
+    posthog_groups: Optional[Dict[str, Any]],
+    base_url: URL,
+    call_method: Callable[..., Any],
+    **kwargs: Any,
+) -> Any:
+    """
+    Common usage-tracking logic for both sync and async calls.
+    call_method: the llm call method (e.g. openai.chat.completions.create)
+    """
+    start_time = time.time()
+    response = None
+    error = None
+    http_status = 200
+    usage: Dict[str, Any] = {}
+
+    try:
+        response = call_method(**kwargs)
+    except Exception as exc:
+        error = exc
+        http_status = getattr(exc, "status_code", 0)  # default to 0 becuase its likely an SDK error
+    finally:
+        end_time = time.time()
+        latency = end_time - start_time
+
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        if response and hasattr(response, "usage"):
+            usage = get_usage(response, provider)
+
+        messages = merge_system_prompt(kwargs, provider)
+
+        event_properties = {
+            "$ai_provider": provider,
+            "$ai_model": kwargs.get("model"),
+            "$ai_model_parameters": get_model_params(kwargs),
+            "$ai_input": with_privacy_mode(ph_client, posthog_privacy_mode, messages),
+            "$ai_output_choices": with_privacy_mode(
+                ph_client, posthog_privacy_mode, format_response(response, provider)
+            ),
+            "$ai_http_status": http_status,
+            "$ai_input_tokens": usage.get("input_tokens", 0),
+            "$ai_output_tokens": usage.get("output_tokens", 0),
+            "$ai_latency": latency,
+            "$ai_trace_id": posthog_trace_id,
+            "$ai_base_url": str(base_url),
+            **(posthog_properties or {}),
+        }
+
+        if posthog_distinct_id is None:
+            event_properties["$process_person_profile"] = False
+
+        # send the event to posthog
+        if hasattr(ph_client, "capture") and callable(ph_client.capture):
+            ph_client.capture(
+                distinct_id=posthog_distinct_id or posthog_trace_id,
+                event="$ai_generation",
+                properties=event_properties,
+                groups=posthog_groups,
+            )
+
+    if error:
+        raise error
+
+    return response
+
+
+async def call_llm_and_track_usage_async(
+    posthog_distinct_id: Optional[str],
+    ph_client: PostHogClient,
+    provider: str,
+    posthog_trace_id: Optional[str],
+    posthog_properties: Optional[Dict[str, Any]],
+    posthog_privacy_mode: bool,
+    posthog_groups: Optional[Dict[str, Any]],
+    base_url: URL,
+    call_async_method: Callable[..., Any],
+    **kwargs: Any,
+) -> Any:
+    start_time = time.time()
+    response = None
+    error = None
+    http_status = 200
+    usage: Dict[str, Any] = {}
+
+    try:
+        response = await call_async_method(**kwargs)
+    except Exception as exc:
+        error = exc
+        http_status = getattr(exc, "status_code", 0)  # default to 0 because its likely an SDK error
+    finally:
+        end_time = time.time()
+        latency = end_time - start_time
+
+        if posthog_trace_id is None:
+            posthog_trace_id = uuid.uuid4()
+
+        if response and hasattr(response, "usage"):
+            usage = get_usage(response, provider)
+
+        messages = merge_system_prompt(kwargs, provider)
+
+        event_properties = {
+            "$ai_provider": provider,
+            "$ai_model": kwargs.get("model"),
+            "$ai_model_parameters": get_model_params(kwargs),
+            "$ai_input": with_privacy_mode(ph_client, posthog_privacy_mode, messages),
+            "$ai_output_choices": with_privacy_mode(
+                ph_client, posthog_privacy_mode, format_response(response, provider)
+            ),
+            "$ai_http_status": http_status,
+            "$ai_input_tokens": usage.get("input_tokens", 0),
+            "$ai_output_tokens": usage.get("output_tokens", 0),
+            "$ai_latency": latency,
+            "$ai_trace_id": posthog_trace_id,
+            "$ai_base_url": str(base_url),
+            **(posthog_properties or {}),
+        }
+
+        if posthog_distinct_id is None:
+            event_properties["$process_person_profile"] = False
+
+        # send the event to posthog
+        if hasattr(ph_client, "capture") and callable(ph_client.capture):
+            ph_client.capture(
+                distinct_id=posthog_distinct_id or posthog_trace_id,
+                event="$ai_generation",
+                properties=event_properties,
+                groups=posthog_groups,
+            )
+
+    if error:
+        raise error
+
+    return response
+
+
+def with_privacy_mode(ph_client: PostHogClient, privacy_mode: bool, value: Any):
+    if ph_client.privacy_mode or privacy_mode:
+        return None
+    return value

posthog/client.py

@@ -1,17 +1,21 @@
 import atexit
 import logging
 import numbers
+import os
+import sys
 from datetime import datetime, timedelta
-from uuid import UUID
+from uuid import UUID, uuid4
 
 from dateutil.tz import tzutc
 from six import string_types
 
 from posthog.consumer import Consumer
+from posthog.exception_capture import ExceptionCapture
+from posthog.exception_utils import exc_info_from_error, exceptions_from_error_tuple, handle_in_app
 from posthog.feature_flags import InconclusiveMatchError, match_feature_flag_properties
 from posthog.poller import Poller
-from posthog.request import APIError, batch_post, decide, get
-from posthog.utils import SizeLimitedDict, clean, guess_timezone
+from posthog.request import DEFAULT_HOST, APIError, batch_post, decide, determine_server_host, get
+from posthog.utils import SizeLimitedDict, clean, guess_timezone, remove_trailing_slash
 from posthog.version import VERSION
 
 try:
@@ -49,6 +53,13 @@ class Client(object):
         project_api_key=None,
         disabled=False,
         disable_geoip=True,
+        historical_migration=False,
+        feature_flags_request_timeout_seconds=3,
+        super_properties=None,
+        enable_exception_autocapture=False,
+        exception_autocapture_integrations=None,
+        project_root=None,
+        privacy_mode=False,
     ):
         self.queue = queue.Queue(max_queue_size)
 
@@ -61,7 +72,9 @@ class Client(object):
         self.debug = debug
         self.send = send
         self.sync_mode = sync_mode
-        self.host = host
+        # Used for session replay URL generation - we don't want the server host here.
+        self.raw_host = host or DEFAULT_HOST
+        self.host = determine_server_host(host)
         self.gzip = gzip
         self.timeout = timeout
         self.feature_flags = None
@@ -69,10 +82,25 @@ class Client(object):
         self.group_type_mapping = None
         self.cohorts = None
         self.poll_interval = poll_interval
+        self.feature_flags_request_timeout_seconds = feature_flags_request_timeout_seconds
         self.poller = None
         self.distinct_ids_feature_flags_reported = SizeLimitedDict(MAX_DICT_SIZE, set)
         self.disabled = disabled
         self.disable_geoip = disable_geoip
+        self.historical_migration = historical_migration
+        self.super_properties = super_properties
+        self.enable_exception_autocapture = enable_exception_autocapture
+        self.exception_autocapture_integrations = exception_autocapture_integrations
+        self.exception_capture = None
+        self.privacy_mode = privacy_mode
+
+        if project_root is None:
+            try:
+                project_root = os.getcwd()
+            except Exception:
+                project_root = None
+
+        self.project_root = project_root
 
         # personal_api_key: This should be a generated Personal API Key, private
         self.personal_api_key = personal_api_key
@@ -84,6 +112,9 @@ class Client(object):
         else:
             self.log.setLevel(logging.WARNING)
 
+        if self.enable_exception_autocapture:
+            self.exception_capture = ExceptionCapture(self, integrations=self.exception_autocapture_integrations)
+
         if sync_mode:
             self.consumers = None
         else:
@@ -100,13 +131,14 @@ class Client(object):
                 consumer = Consumer(
                     self.queue,
                     self.api_key,
-                    host=host,
+                    host=self.host,
                     on_error=on_error,
                     flush_at=flush_at,
                     flush_interval=flush_interval,
                     gzip=gzip,
                     retries=max_retries,
                     timeout=timeout,
+                    historical_migration=historical_migration,
                 )
                 self.consumers.append(consumer)
 
@@ -143,6 +175,15 @@ class Client(object):
         resp_data = self.get_decide(distinct_id, groups, person_properties, group_properties, disable_geoip)
         return resp_data["featureFlagPayloads"]
 
+    def get_feature_flags_and_payloads(
+        self, distinct_id, groups=None, person_properties=None, group_properties=None, disable_geoip=None
+    ):
+        resp_data = self.get_decide(distinct_id, groups, person_properties, group_properties, disable_geoip)
+        return {
+            "featureFlags": resp_data["featureFlags"],
+            "featureFlagPayloads": resp_data["featureFlagPayloads"],
+        }
+
     def get_decide(self, distinct_id, groups=None, person_properties=None, group_properties=None, disable_geoip=None):
         require("distinct_id", distinct_id, ID_TYPES)
 
@@ -161,7 +202,7 @@ class Client(object):
             "group_properties": group_properties,
             "disable_geoip": disable_geoip,
         }
-        resp_data = decide(self.api_key, self.host, timeout=10, **request_data)
+        resp_data = decide(self.api_key, self.host, timeout=self.feature_flags_request_timeout_seconds, **request_data)
 
         return resp_data
 
@@ -265,6 +306,7 @@ class Client(object):
         timestamp=None,
         uuid=None,
         disable_geoip=None,
+        distinct_id=None,
     ):
         properties = properties or {}
         context = context or {}
@@ -272,6 +314,11 @@ class Client(object):
         require("group_key", group_key, ID_TYPES)
         require("properties", properties, dict)
 
+        if distinct_id:
+            require("distinct_id", distinct_id, ID_TYPES)
+        else:
+            distinct_id = "${}_{}".format(group_type, group_key)
+
         msg = {
             "event": "$groupidentify",
             "properties": {
@@ -279,7 +326,7 @@ class Client(object):
                 "$group_key": group_key,
                 "$group_set": properties,
             },
-            "distinct_id": "${}_{}".format(group_type, group_key),
+            "distinct_id": distinct_id,
             "timestamp": timestamp,
             "context": context,
             "uuid": uuid,
@@ -329,6 +376,65 @@ class Client(object):
 
         return self._enqueue(msg, disable_geoip)
 
+    def capture_exception(
+        self,
+        exception=None,
+        distinct_id=None,
+        properties=None,
+        context=None,
+        timestamp=None,
+        uuid=None,
+        groups=None,
+    ):
+        # this function shouldn't ever throw an error, so it logs exceptions instead of raising them.
+        # this is important to ensure we don't unexpectedly re-raise exceptions in the user's code.
+        try:
+            properties = properties or {}
+
+            # if there's no distinct_id, we'll generate one and set personless mode
+            # via $process_person_profile = false
+            if distinct_id is None:
+                properties["$process_person_profile"] = False
+                distinct_id = uuid4()
+
+            require("distinct_id", distinct_id, ID_TYPES)
+            require("properties", properties, dict)
+
+            if exception is not None:
+                exc_info = exc_info_from_error(exception)
+            else:
+                exc_info = sys.exc_info()
+
+            if exc_info is None or exc_info == (None, None, None):
+                self.log.warning("No exception information available")
+                return
+
+            # Format stack trace for cymbal
+            all_exceptions_with_trace = exceptions_from_error_tuple(exc_info)
+
+            # Add in-app property to frames in the exceptions
+            event = handle_in_app(
+                {
+                    "exception": {
+                        "values": all_exceptions_with_trace,
+                    },
+                },
+                project_root=self.project_root,
+            )
+            all_exceptions_with_trace_and_in_app = event["exception"]["values"]
+
+            properties = {
+                "$exception_type": all_exceptions_with_trace_and_in_app[0].get("type"),
+                "$exception_message": all_exceptions_with_trace_and_in_app[0].get("value"),
+                "$exception_list": all_exceptions_with_trace_and_in_app,
+                "$exception_personURL": f"{remove_trailing_slash(self.raw_host)}/project/{self.api_key}/person/{distinct_id}",
+                **properties,
+            }
+
+            return self.capture(distinct_id, "$exception", properties, context, timestamp, uuid, groups)
+        except Exception as e:
+            self.log.exception(f"Failed to capture exception: {e}")
+
     def _enqueue(self, msg, disable_geoip):
         """Push a new `msg` onto the queue, return `(success, msg)`"""
 
@@ -337,7 +443,7 @@ class Client(object):
 
         timestamp = msg["timestamp"]
         if timestamp is None:
-            timestamp = datetime.utcnow().replace(tzinfo=tzutc())
+            timestamp = datetime.now(tz=tzutc())
 
         require("timestamp", timestamp, datetime)
         require("context", msg["context"], dict)
@@ -363,6 +469,9 @@ class Client(object):
         if disable_geoip:
             msg["properties"]["$geoip_disable"] = True
 
+        if self.super_properties:
+            msg["properties"] = {**msg["properties"], **self.super_properties}
+
         msg["distinct_id"] = stringify_id(msg.get("distinct_id", None))
 
         msg = clean(msg)
@@ -374,7 +483,14 @@ class Client(object):
 
         if self.sync_mode:
             self.log.debug("enqueued with blocking %s.", msg["event"])
-            batch_post(self.api_key, self.host, gzip=self.gzip, timeout=self.timeout, batch=[msg])
+            batch_post(
+                self.api_key,
+                self.host,
+                gzip=self.gzip,
+                timeout=self.timeout,
+                batch=[msg],
+                historical_migration=self.historical_migration,
+            )
 
             return True, msg
 
@@ -414,6 +530,9 @@ class Client(object):
         self.flush()
         self.join()
 
+        if self.exception_capture:
+            self.exception_capture.close()
+
     def _load_feature_flags(self):
         try:
             response = get(
@@ -451,7 +570,7 @@ class Client(object):
             )
             self.log.warning(e)
 
-        self._last_feature_flag_poll = datetime.utcnow().replace(tzinfo=tzutc())
+        self._last_feature_flag_poll = datetime.now(tz=tzutc())
 
     def load_feature_flags(self):
         if not self.personal_api_key:
@@ -464,7 +583,16 @@ class Client(object):
             self.poller = Poller(interval=timedelta(seconds=self.poll_interval), execute=self._load_feature_flags)
             self.poller.start()
 
-    def _compute_flag_locally(self, feature_flag, distinct_id, *, groups={}, person_properties={}, group_properties={}):
+    def _compute_flag_locally(
+        self,
+        feature_flag,
+        distinct_id,
+        *,
+        groups={},
+        person_properties={},
+        group_properties={},
+        warn_on_unknown_groups=True,
+    ):
         if feature_flag.get("ensure_experience_continuity", False):
             raise InconclusiveMatchError("Flag has experience continuity enabled")
 
@@ -486,9 +614,14 @@ class Client(object):
             if group_name not in groups:
                 # Group flags are never enabled in `groups` aren't passed in
                 # don't failover to `/decide/`, since response will be the same
-                self.log.warning(
-                    f"[FEATURE FLAGS] Can't compute group feature flag: {feature_flag['key']} without group names passed in"
-                )
+                if warn_on_unknown_groups:
+                    self.log.warning(
+                        f"[FEATURE FLAGS] Can't compute group feature flag: {feature_flag['key']} without group names passed in"
+                    )
+                else:
+                    self.log.debug(
+                        f"[FEATURE FLAGS] Can't compute group feature flag: {feature_flag['key']} without group names passed in"
+                    )
                 return False
 
             focused_group_properties = group_properties[group_name]
@@ -630,23 +763,52 @@ class Client(object):
                 groups=groups,
                 person_properties=person_properties,
                 group_properties=group_properties,
-                send_feature_flag_events=send_feature_flag_events,
-                only_evaluate_locally=True,
+                send_feature_flag_events=False,
+                # Disable automatic sending of feature flag events because we're manually handling event dispatch.
+                # This prevents sending events with empty data when `get_feature_flag` cannot be evaluated locally.
+                only_evaluate_locally=True,  # Enable local evaluation of feature flags to avoid making multiple requests to `/decide`.
                 disable_geoip=disable_geoip,
             )
 
         response = None
+        payload = None
 
         if match_value is not None:
-            response = self._compute_payload_locally(key, match_value)
+            payload = self._compute_payload_locally(key, match_value)
 
-        if response is None and not only_evaluate_locally:
-            decide_payloads = self.get_feature_payloads(
-                distinct_id, groups, person_properties, group_properties, disable_geoip
+        flag_was_locally_evaluated = payload is not None
+        if not flag_was_locally_evaluated and not only_evaluate_locally:
+            try:
+                responses_and_payloads = self.get_feature_flags_and_payloads(
+                    distinct_id, groups, person_properties, group_properties, disable_geoip
+                )
+                response = responses_and_payloads["featureFlags"].get(key, None)
+                payload = responses_and_payloads["featureFlagPayloads"].get(str(key).lower(), None)
+            except Exception as e:
+                self.log.exception(f"[FEATURE FLAGS] Unable to get feature flags and payloads: {e}")
+
+        feature_flag_reported_key = f"{key}_{str(response)}"
+
+        if (
+            feature_flag_reported_key not in self.distinct_ids_feature_flags_reported[distinct_id]
+            and send_feature_flag_events  # noqa: W503
+        ):
+            self.capture(
+                distinct_id,
+                "$feature_flag_called",
+                {
+                    "$feature_flag": key,
+                    "$feature_flag_response": response,
+                    "$feature_flag_payload": payload,
+                    "locally_evaluated": flag_was_locally_evaluated,
+                    f"$feature/{key}": response,
+                },
+                groups=groups,
+                disable_geoip=disable_geoip,
             )
-            response = decide_payloads.get(str(key).lower(), None)
+            self.distinct_ids_feature_flags_reported[distinct_id].add(feature_flag_reported_key)
 
-        return response
+        return payload
 
     def _compute_payload_locally(self, key, match_value):
         payload = None
@@ -717,7 +879,9 @@ class Client(object):
 
         return response
 
-    def _get_all_flags_and_payloads_locally(self, distinct_id, *, groups={}, person_properties={}, group_properties={}):
+    def _get_all_flags_and_payloads_locally(
+        self, distinct_id, *, groups={}, person_properties={}, group_properties={}, warn_on_unknown_groups=False
+    ):
         require("distinct_id", distinct_id, ID_TYPES)
         require("groups", groups, dict)
 
@@ -737,6 +901,7 @@ class Client(object):
                         groups=groups,
                         person_properties=person_properties,
                         group_properties=group_properties,
+                        warn_on_unknown_groups=warn_on_unknown_groups,
                     )
                     matched_payload = self._compute_payload_locally(flag["key"], flags[flag["key"]])
                     if matched_payload:
@@ -756,7 +921,7 @@ class Client(object):
         return self.feature_flags
 
     def _add_local_person_and_group_properties(self, distinct_id, groups, person_properties, group_properties):
-        all_person_properties = {"$current_distinct_id": distinct_id, **(person_properties or {})}
+        all_person_properties = {"distinct_id": distinct_id, **(person_properties or {})}
 
         all_group_properties = {}
         if groups:

posthog/consumer.py

@@ -36,6 +36,7 @@ class Consumer(Thread):
         gzip=False,
         retries=10,
         timeout=15,
+        historical_migration=False,
     ):
         """Create a consumer thread."""
         Thread.__init__(self)
@@ -55,6 +56,7 @@ class Consumer(Thread):
         self.running = True
         self.retries = retries
         self.timeout = timeout
+        self.historical_migration = historical_migration
 
     def run(self):
         """Runs the consumer."""
@@ -134,6 +136,13 @@ class Consumer(Thread):
 
         @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)
+            batch_post(
+                self.api_key,
+                self.host,
+                gzip=self.gzip,
+                timeout=self.timeout,
+                batch=batch,
+                historical_migration=self.historical_migration,
+            )
 
         send_request()

posthog/exception_capture.py

@@ -0,0 +1,64 @@
+import logging
+import sys
+import threading
+from enum import Enum
+from typing import TYPE_CHECKING, List, Optional
+
+if TYPE_CHECKING:
+    from posthog.client import Client
+
+
+class Integrations(str, Enum):
+    Django = "django"
+
+
+class ExceptionCapture:
+    # TODO: Add client side rate limiting to prevent spamming the server with exceptions
+
+    log = logging.getLogger("posthog")
+
+    def __init__(self, client: "Client", integrations: Optional[List[Integrations]] = None):
+        self.client = client
+        self.original_excepthook = sys.excepthook
+        sys.excepthook = self.exception_handler
+        threading.excepthook = self.thread_exception_handler
+        self.enabled_integrations = []
+
+        for integration in integrations or []:
+            # TODO: Maybe find a better way of enabling integrations
+            # This is very annoying currently if we had to add any configuration per integration
+            if integration == Integrations.Django:
+                try:
+                    from posthog.exception_integrations.django import DjangoIntegration
+
+                    enabled_integration = DjangoIntegration(self.exception_receiver)
+                    self.enabled_integrations.append(enabled_integration)
+                except Exception as e:
+                    self.log.exception(f"Failed to enable Django integration: {e}")
+
+    def close(self):
+        sys.excepthook = self.original_excepthook
+        for integration in self.enabled_integrations:
+            integration.uninstall()
+
+    def exception_handler(self, exc_type, exc_value, exc_traceback):
+        # don't affect default behaviour.
+        self.capture_exception((exc_type, exc_value, exc_traceback))
+        self.original_excepthook(exc_type, exc_value, exc_traceback)
+
+    def thread_exception_handler(self, args):
+        self.capture_exception((args.exc_type, args.exc_value, args.exc_traceback))
+
+    def exception_receiver(self, exc_info, extra_properties):
+        if "distinct_id" in extra_properties:
+            metadata = {"distinct_id": extra_properties["distinct_id"]}
+        else:
+            metadata = None
+        self.capture_exception((exc_info[0], exc_info[1], exc_info[2]), metadata)
+
+    def capture_exception(self, exception, metadata=None):
+        try:
+            distinct_id = metadata.get("distinct_id") if metadata else None
+            self.client.capture_exception(exception, distinct_id)
+        except Exception as e:
+            self.log.exception(f"Failed to capture exception: {e}")

posthog/exception_integrations/__init__.py

@@ -0,0 +1,5 @@
+class IntegrationEnablingError(Exception):
+    """
+    The integration could not be enabled due to a user error like
+    `django` not being installed for the `DjangoIntegration`.
+    """

posthog/exception_integrations/django.py

@@ -0,0 +1,88 @@
+import re
+import sys
+from typing import TYPE_CHECKING
+
+from posthog.exception_integrations import IntegrationEnablingError
+
+try:
+    from django import VERSION as DJANGO_VERSION
+    from django.core import signals
+
+except ImportError:
+    raise IntegrationEnablingError("Django not installed")
+
+
+if TYPE_CHECKING:
+    from typing import Any, Dict  # noqa: F401
+
+    from django.core.handlers.wsgi import WSGIRequest  # noqa: F401
+
+
+class DjangoIntegration:
+    # TODO: Abstract integrations one we have more and can see patterns
+    """
+    Autocapture errors from a Django application.
+    """
+
+    identifier = "django"
+
+    def __init__(self, capture_exception_fn=None):
+
+        if DJANGO_VERSION < (4, 2):
+            raise IntegrationEnablingError("Django 4.2 or newer is required.")
+
+        # TODO: Right now this seems too complicated / overkill for us, but seems like we can automatically plug in middlewares
+        # which is great for users (they don't need to do this) and everything should just work.
+        # We should consider this in the future, but for now we can just use the middleware and signals handlers.
+        # See: https://github.com/getsentry/sentry-python/blob/269d96d6e9821122fbff280e6a26956e5ed03c0b/sentry_sdk/integrations/django/__init__.py
+
+        self.capture_exception_fn = capture_exception_fn
+
+        def _got_request_exception(request=None, **kwargs):
+            # type: (WSGIRequest, **Any) -> None
+
+            extra_props = {}
+            if request is not None:
+                # get headers metadata
+                extra_props = DjangoRequestExtractor(request).extract_person_data()
+
+            self.capture_exception_fn(sys.exc_info(), extra_props)
+
+        signals.got_request_exception.connect(_got_request_exception)
+
+    def uninstall(self):
+        pass
+
+
+class DjangoRequestExtractor:
+
+    def __init__(self, request):
+        # type: (Any) -> None
+        self.request = request
+
+    def extract_person_data(self):
+        headers = self.headers()
+
+        # Extract traceparent and tracestate headers
+        traceparent = headers.get("traceparent")
+        tracestate = headers.get("tracestate")
+
+        # Extract the distinct_id from tracestate
+        distinct_id = None
+        if tracestate:
+            # TODO: Align on the format of the distinct_id in tracestate
+            # We can't have comma or equals in header values here, so maybe we should base64 encode it?
+            match = re.search(r"posthog-distinct-id=([^,]+)", tracestate)
+            if match:
+                distinct_id = match.group(1)
+
+        return {
+            "distinct_id": distinct_id,
+            "ip": headers.get("X-Forwarded-For"),
+            "user_agent": headers.get("User-Agent"),
+            "traceparent": traceparent,
+        }
+
+    def headers(self):
+        # type: () -> Dict[str, str]
+        return dict(self.request.headers)

posthog/exception_utils.py

@@ -0,0 +1,873 @@
+# copied and adapted from https://github.com/getsentry/sentry-python/blob/269d96d6e9821122fbff280e6a26956e5ed03c0b/sentry_sdk/utils.py#L689
+# 💖open source (under MIT License)
+# We want to keep payloads as similar to Sentry as possible for easy interoperability
+
+import linecache
+import os
+import re
+import sys
+from datetime import datetime
+from typing import TYPE_CHECKING
+
+try:
+    # Python 3.11
+    from builtins import BaseExceptionGroup
+except ImportError:
+    # Python 3.10 and below
+    BaseExceptionGroup = None  # type: ignore
+
+
+DEFAULT_MAX_VALUE_LENGTH = 1024
+
+
+if TYPE_CHECKING:
+
+    from types import FrameType, TracebackType
+    from typing import (  # noqa: F401
+        Any,
+        Callable,
+        Dict,
+        Iterator,
+        List,
+        Literal,
+        Optional,
+        Set,
+        Tuple,
+        Type,
+        TypedDict,
+        TypeVar,
+        Union,
+        cast,
+    )
+
+    ExcInfo = Union[
+        Tuple[Type[BaseException], BaseException, Optional[TracebackType]],
+        Tuple[None, None, None],
+    ]
+    LogLevelStr = Literal["fatal", "critical", "error", "warning", "info", "debug"]
+
+    Event = TypedDict(
+        "Event",
+        {
+            "breadcrumbs": Dict[Literal["values"], List[Dict[str, Any]]],  # TODO: We can expand on this type
+            "check_in_id": str,
+            "contexts": Dict[str, Dict[str, object]],
+            "dist": str,
+            "duration": Optional[float],
+            "environment": str,
+            "errors": List[Dict[str, Any]],  # TODO: We can expand on this type
+            "event_id": str,
+            "exception": Dict[Literal["values"], List[Dict[str, Any]]],  # TODO: We can expand on this type
+            # "extra": MutableMapping[str, object],
+            # "fingerprint": List[str],
+            "level": LogLevelStr,
+            # "logentry": Mapping[str, object],
+            "logger": str,
+            # "measurements": Dict[str, MeasurementValue],
+            "message": str,
+            "modules": Dict[str, str],
+            # "monitor_config": Mapping[str, object],
+            "monitor_slug": Optional[str],
+            "platform": Literal["python"],
+            "profile": object,  # Should be sentry_sdk.profiler.Profile, but we can't import that here due to circular imports
+            "release": str,
+            "request": Dict[str, object],
+            # "sdk": Mapping[str, object],
+            "server_name": str,
+            "spans": List[Dict[str, object]],
+            "stacktrace": Dict[str, object],  # We access this key in the code, but I am unsure whether we ever set it
+            "start_timestamp": datetime,
+            "status": Optional[str],
+            # "tags": MutableMapping[
+            #     str, str
+            # ],  # Tags must be less than 200 characters each
+            "threads": Dict[Literal["values"], List[Dict[str, Any]]],  # TODO: We can expand on this type
+            "timestamp": Optional[datetime],  # Must be set before sending the event
+            "transaction": str,
+            # "transaction_info": Mapping[str, Any],  # TODO: We can expand on this type
+            "type": Literal["check_in", "transaction"],
+            "user": Dict[str, object],
+            "_metrics_summary": Dict[str, object],
+        },
+        total=False,
+    )
+
+
+epoch = datetime(1970, 1, 1)
+
+
+BASE64_ALPHABET = re.compile(r"^[a-zA-Z0-9/+=]*$")
+
+SENSITIVE_DATA_SUBSTITUTE = "[Filtered]"
+
+
+def to_timestamp(value):
+    # type: (datetime) -> float
+    return (value - epoch).total_seconds()
+
+
+def format_timestamp(value):
+    # type: (datetime) -> str
+    return value.strftime("%Y-%m-%dT%H:%M:%S.%fZ")
+
+
+def event_hint_with_exc_info(exc_info=None):
+    # type: (Optional[ExcInfo]) -> Dict[str, Optional[ExcInfo]]
+    """Creates a hint with the exc info filled in."""
+    if exc_info is None:
+        exc_info = sys.exc_info()
+    else:
+        exc_info = exc_info_from_error(exc_info)
+    if exc_info[0] is None:
+        exc_info = None
+    return {"exc_info": exc_info}
+
+
+class AnnotatedValue:
+    """
+    Meta information for a data field in the event payload.
+    This is to tell Relay that we have tampered with the fields value.
+    See:
+    https://github.com/getsentry/relay/blob/be12cd49a0f06ea932ed9b9f93a655de5d6ad6d1/relay-general/src/types/meta.rs#L407-L423
+    """
+
+    __slots__ = ("value", "metadata")
+
+    def __init__(self, value, metadata):
+        # type: (Optional[Any], Dict[str, Any]) -> None
+        self.value = value
+        self.metadata = metadata
+
+    def __eq__(self, other):
+        # type: (Any) -> bool
+        if not isinstance(other, AnnotatedValue):
+            return False
+
+        return self.value == other.value and self.metadata == other.metadata
+
+    @classmethod
+    def removed_because_raw_data(cls):
+        # type: () -> AnnotatedValue
+        """The value was removed because it could not be parsed. This is done for request body values that are not json nor a form."""
+        return AnnotatedValue(
+            value="",
+            metadata={
+                "rem": [  # Remark
+                    [
+                        "!raw",  # Unparsable raw data
+                        "x",  # The fields original value was removed
+                    ]
+                ]
+            },
+        )
+
+    @classmethod
+    def removed_because_over_size_limit(cls):
+        # type: () -> AnnotatedValue
+        """The actual value was removed because the size of the field exceeded the configured maximum size (specified with the max_request_body_size sdk option)"""
+        return AnnotatedValue(
+            value="",
+            metadata={
+                "rem": [  # Remark
+                    [
+                        "!config",  # Because of configured maximum size
+                        "x",  # The fields original value was removed
+                    ]
+                ]
+            },
+        )
+
+    @classmethod
+    def substituted_because_contains_sensitive_data(cls):
+        # type: () -> AnnotatedValue
+        """The actual value was removed because it contained sensitive information."""
+        return AnnotatedValue(
+            value=SENSITIVE_DATA_SUBSTITUTE,
+            metadata={
+                "rem": [  # Remark
+                    [
+                        "!config",  # Because of SDK configuration (in this case the config is the hard coded removal of certain django cookies)
+                        "s",  # The fields original value was substituted
+                    ]
+                ]
+            },
+        )
+
+
+if TYPE_CHECKING:
+    T = TypeVar("T")
+    Annotated = Union[AnnotatedValue, T]
+
+
+def get_type_name(cls):
+    # type: (Optional[type]) -> Optional[str]
+    return getattr(cls, "__qualname__", None) or getattr(cls, "__name__", None)
+
+
+def get_type_module(cls):
+    # type: (Optional[type]) -> Optional[str]
+    mod = getattr(cls, "__module__", None)
+    if mod not in (None, "builtins", "__builtins__"):
+        return mod
+    return None
+
+
+def should_hide_frame(frame: "FrameType") -> bool:
+    try:
+        mod = frame.f_globals["__name__"]
+        if mod.startswith("sentry_sdk."):
+            return True
+    except (AttributeError, KeyError):
+        pass
+
+    for flag_name in "__traceback_hide__", "__tracebackhide__":
+        try:
+            if frame.f_locals[flag_name]:
+                return True
+        except Exception:
+            pass
+
+    return False
+
+
+def iter_stacks(tb):
+    # type: (Optional[TracebackType]) -> Iterator[TracebackType]
+    tb_ = tb  # type: Optional[TracebackType]
+    while tb_ is not None:
+        if not should_hide_frame(tb_.tb_frame):
+            yield tb_
+        tb_ = tb_.tb_next
+
+
+def get_lines_from_file(
+    filename,  # type: str
+    lineno,  # type: int
+    max_length=None,  # type: Optional[int]
+    loader=None,  # type: Optional[Any]
+    module=None,  # type: Optional[str]
+):
+    # type: (...) -> Tuple[List[Annotated[str]], Optional[Annotated[str]], List[Annotated[str]]]
+    context_lines = 5
+    source = None
+    if loader is not None and hasattr(loader, "get_source"):
+        try:
+            source_str = loader.get_source(module)  # type: Optional[str]
+        except (ImportError, IOError):
+            source_str = None
+        if source_str is not None:
+            source = source_str.splitlines()
+
+    if source is None:
+        try:
+            source = linecache.getlines(filename)
+        except (OSError, IOError):
+            return [], None, []
+
+    if not source:
+        return [], None, []
+
+    lower_bound = max(0, lineno - context_lines)
+    upper_bound = min(lineno + 1 + context_lines, len(source))
+
+    try:
+        pre_context = [strip_string(line.strip("\r\n"), max_length=max_length) for line in source[lower_bound:lineno]]
+        context_line = strip_string(source[lineno].strip("\r\n"), max_length=max_length)
+        post_context = [
+            strip_string(line.strip("\r\n"), max_length=max_length)
+            for line in source[(lineno + 1) : upper_bound]  # noqa: E203
+        ]
+        return pre_context, context_line, post_context
+    except IndexError:
+        # the file may have changed since it was loaded into memory
+        return [], None, []
+
+
+def get_source_context(
+    frame,  # type: FrameType
+    tb_lineno,  # type: int
+    max_value_length=None,  # type: Optional[int]
+):
+    # type: (...) -> Tuple[List[Annotated[str]], Optional[Annotated[str]], List[Annotated[str]]]
+    try:
+        abs_path = frame.f_code.co_filename  # type: Optional[str]
+    except Exception:
+        abs_path = None
+    try:
+        module = frame.f_globals["__name__"]
+    except Exception:
+        return [], None, []
+    try:
+        loader = frame.f_globals["__loader__"]
+    except Exception:
+        loader = None
+    lineno = tb_lineno - 1
+    if lineno is not None and abs_path:
+        return get_lines_from_file(abs_path, lineno, max_value_length, loader=loader, module=module)
+    return [], None, []
+
+
+def safe_str(value):
+    # type: (Any) -> str
+    try:
+        return str(value)
+    except Exception:
+        return safe_repr(value)
+
+
+def safe_repr(value):
+    # type: (Any) -> str
+    try:
+        return repr(value)
+    except Exception:
+        return "<broken repr>"
+
+
+def filename_for_module(module, abs_path):
+    # type: (Optional[str], Optional[str]) -> Optional[str]
+    if not abs_path or not module:
+        return abs_path
+
+    try:
+        if abs_path.endswith(".pyc"):
+            abs_path = abs_path[:-1]
+
+        base_module = module.split(".", 1)[0]
+        if base_module == module:
+            return os.path.basename(abs_path)
+
+        base_module_path = sys.modules[base_module].__file__
+        if not base_module_path:
+            return abs_path
+
+        return abs_path.split(base_module_path.rsplit(os.sep, 2)[0], 1)[-1].lstrip(os.sep)
+    except Exception:
+        return abs_path
+
+
+def serialize_frame(
+    frame,
+    tb_lineno=None,
+    include_local_variables=True,
+    include_source_context=True,
+    max_value_length=None,
+    custom_repr=None,
+):
+    # type: (FrameType, Optional[int], bool, bool, Optional[int], Optional[Callable[..., Optional[str]]]) -> Dict[str, Any]
+    f_code = getattr(frame, "f_code", None)
+    if not f_code:
+        abs_path = None
+        function = None
+    else:
+        abs_path = frame.f_code.co_filename
+        function = frame.f_code.co_name
+    try:
+        module = frame.f_globals["__name__"]
+    except Exception:
+        module = None
+
+    if tb_lineno is None:
+        tb_lineno = frame.f_lineno
+
+    rv = {
+        "platform": "python",
+        "filename": filename_for_module(module, abs_path) or None,
+        "abs_path": os.path.abspath(abs_path) if abs_path else None,
+        "function": function or "<unknown>",
+        "module": module,
+        "lineno": tb_lineno,
+    }  # type: Dict[str, Any]
+
+    if include_source_context:
+        rv["pre_context"], rv["context_line"], rv["post_context"] = get_source_context(
+            frame, tb_lineno, max_value_length
+        )
+
+    if include_local_variables:
+        # TODO(nk): Sort out this current invalid import
+        # from sentry_sdk.serializer import serialize
+
+        # rv["vars"] = serialize(
+        #     dict(frame.f_locals), is_vars=True, custom_repr=custom_repr
+        # )
+        pass
+
+    return rv
+
+
+def current_stacktrace(
+    include_local_variables=True,  # type: bool
+    include_source_context=True,  # type: bool
+    max_value_length=None,  # type: Optional[int]
+):
+    # type: (...) -> Dict[str, Any]
+    __tracebackhide__ = True
+    frames = []
+
+    f = sys._getframe()  # type: Optional[FrameType]
+    while f is not None:
+        if not should_hide_frame(f):
+            frames.append(
+                serialize_frame(
+                    f,
+                    include_local_variables=include_local_variables,
+                    include_source_context=include_source_context,
+                    max_value_length=max_value_length,
+                )
+            )
+        f = f.f_back
+
+    frames.reverse()
+
+    return {"frames": frames, "type": "raw"}
+
+
+def get_errno(exc_value):
+    # type: (BaseException) -> Optional[Any]
+    return getattr(exc_value, "errno", None)
+
+
+def get_error_message(exc_value):
+    # type: (Optional[BaseException]) -> str
+    return getattr(exc_value, "message", "") or getattr(exc_value, "detail", "") or safe_str(exc_value)
+
+
+def single_exception_from_error_tuple(
+    exc_type,  # type: Optional[type]
+    exc_value,  # type: Optional[BaseException]
+    tb,  # type: Optional[TracebackType]
+    client_options=None,  # type: Optional[Dict[str, Any]]
+    mechanism=None,  # type: Optional[Dict[str, Any]]
+    exception_id=None,  # type: Optional[int]
+    parent_id=None,  # type: Optional[int]
+    source=None,  # type: Optional[str]
+):
+    # type: (...) -> Dict[str, Any]
+    """
+    Creates a dict that goes into the events `exception.values` list and is ingestible by Sentry.
+
+    See the Exception Interface documentation for more details:
+    https://develop.sentry.dev/sdk/event-payloads/exception/
+    """
+    exception_value = {}  # type: Dict[str, Any]
+    exception_value["mechanism"] = mechanism.copy() if mechanism else {"type": "generic", "handled": True}
+    if exception_id is not None:
+        exception_value["mechanism"]["exception_id"] = exception_id
+
+    if exc_value is not None:
+        errno = get_errno(exc_value)
+    else:
+        errno = None
+
+    if errno is not None:
+        exception_value["mechanism"].setdefault("meta", {}).setdefault("errno", {}).setdefault("number", errno)
+
+    if source is not None:
+        exception_value["mechanism"]["source"] = source
+
+    is_root_exception = exception_id == 0
+    if not is_root_exception and parent_id is not None:
+        exception_value["mechanism"]["parent_id"] = parent_id
+        exception_value["mechanism"]["type"] = "chained"
+
+    if is_root_exception and "type" not in exception_value["mechanism"]:
+        exception_value["mechanism"]["type"] = "generic"
+
+    is_exception_group = BaseExceptionGroup is not None and isinstance(exc_value, BaseExceptionGroup)
+    if is_exception_group:
+        exception_value["mechanism"]["is_exception_group"] = True
+
+    exception_value["module"] = get_type_module(exc_type)
+    exception_value["type"] = get_type_name(exc_type)
+    exception_value["value"] = get_error_message(exc_value)
+
+    if client_options is None:
+        include_local_variables = True
+        include_source_context = True
+        max_value_length = DEFAULT_MAX_VALUE_LENGTH  # fallback
+        custom_repr = None
+    else:
+        include_local_variables = client_options["include_local_variables"]
+        include_source_context = client_options["include_source_context"]
+        max_value_length = client_options["max_value_length"]
+        custom_repr = client_options.get("custom_repr")
+
+    frames = [
+        serialize_frame(
+            tb.tb_frame,
+            tb_lineno=tb.tb_lineno,
+            include_local_variables=include_local_variables,
+            include_source_context=include_source_context,
+            max_value_length=max_value_length,
+            custom_repr=custom_repr,
+        )
+        for tb in iter_stacks(tb)
+    ]
+
+    if frames:
+        exception_value["stacktrace"] = {"frames": frames, "type": "raw"}
+
+    return exception_value
+
+
+HAS_CHAINED_EXCEPTIONS = hasattr(Exception, "__suppress_context__")
+
+if HAS_CHAINED_EXCEPTIONS:
+
+    def walk_exception_chain(exc_info):
+        # type: (ExcInfo) -> Iterator[ExcInfo]
+        exc_type, exc_value, tb = exc_info
+
+        seen_exceptions = []
+        seen_exception_ids = set()  # type: Set[int]
+
+        while exc_type is not None and exc_value is not None and id(exc_value) not in seen_exception_ids:
+            yield exc_type, exc_value, tb
+
+            # Avoid hashing random types we don't know anything
+            # about. Use the list to keep a ref so that the `id` is
+            # not used for another object.
+            seen_exceptions.append(exc_value)
+            seen_exception_ids.add(id(exc_value))
+
+            if exc_value.__suppress_context__:
+                cause = exc_value.__cause__
+            else:
+                cause = exc_value.__context__
+            if cause is None:
+                break
+            exc_type = type(cause)
+            exc_value = cause
+            tb = getattr(cause, "__traceback__", None)
+
+else:
+
+    def walk_exception_chain(exc_info):
+        # type: (ExcInfo) -> Iterator[ExcInfo]
+        yield exc_info
+
+
+def exceptions_from_error(
+    exc_type,  # type: Optional[type]
+    exc_value,  # type: Optional[BaseException]
+    tb,  # type: Optional[TracebackType]
+    client_options=None,  # type: Optional[Dict[str, Any]]
+    mechanism=None,  # type: Optional[Dict[str, Any]]
+    exception_id=0,  # type: int
+    parent_id=0,  # type: int
+    source=None,  # type: Optional[str]
+):
+    # type: (...) -> Tuple[int, List[Dict[str, Any]]]
+    """
+    Creates the list of exceptions.
+    This can include chained exceptions and exceptions from an ExceptionGroup.
+
+    See the Exception Interface documentation for more details:
+    https://develop.sentry.dev/sdk/event-payloads/exception/
+    """
+
+    parent = single_exception_from_error_tuple(
+        exc_type=exc_type,
+        exc_value=exc_value,
+        tb=tb,
+        client_options=client_options,
+        mechanism=mechanism,
+        exception_id=exception_id,
+        parent_id=parent_id,
+        source=source,
+    )
+    exceptions = [parent]
+
+    parent_id = exception_id
+    exception_id += 1
+
+    should_supress_context = hasattr(exc_value, "__suppress_context__") and exc_value.__suppress_context__  # type: ignore
+    if should_supress_context:
+        # Add direct cause.
+        # The field `__cause__` is set when raised with the exception (using the `from` keyword).
+        exception_has_cause = exc_value and hasattr(exc_value, "__cause__") and exc_value.__cause__ is not None
+        if exception_has_cause:
+            cause = exc_value.__cause__  # type: ignore
+            (exception_id, child_exceptions) = exceptions_from_error(
+                exc_type=type(cause),
+                exc_value=cause,
+                tb=getattr(cause, "__traceback__", None),
+                client_options=client_options,
+                mechanism=mechanism,
+                exception_id=exception_id,
+                source="__cause__",
+            )
+            exceptions.extend(child_exceptions)
+
+    else:
+        # Add indirect cause.
+        # The field `__context__` is assigned if another exception occurs while handling the exception.
+        exception_has_content = exc_value and hasattr(exc_value, "__context__") and exc_value.__context__ is not None
+        if exception_has_content:
+            context = exc_value.__context__  # type: ignore
+            (exception_id, child_exceptions) = exceptions_from_error(
+                exc_type=type(context),
+                exc_value=context,
+                tb=getattr(context, "__traceback__", None),
+                client_options=client_options,
+                mechanism=mechanism,
+                exception_id=exception_id,
+                source="__context__",
+            )
+            exceptions.extend(child_exceptions)
+
+    # Add exceptions from an ExceptionGroup.
+    is_exception_group = exc_value and hasattr(exc_value, "exceptions")
+    if is_exception_group:
+        for idx, e in enumerate(exc_value.exceptions):  # type: ignore
+            (exception_id, child_exceptions) = exceptions_from_error(
+                exc_type=type(e),
+                exc_value=e,
+                tb=getattr(e, "__traceback__", None),
+                client_options=client_options,
+                mechanism=mechanism,
+                exception_id=exception_id,
+                parent_id=parent_id,
+                source="exceptions[%s]" % idx,
+            )
+            exceptions.extend(child_exceptions)
+
+    return (exception_id, exceptions)
+
+
+def exceptions_from_error_tuple(
+    exc_info,  # type: ExcInfo
+    client_options=None,  # type: Optional[Dict[str, Any]]
+    mechanism=None,  # type: Optional[Dict[str, Any]]
+):
+    # type: (...) -> List[Dict[str, Any]]
+    exc_type, exc_value, tb = exc_info
+
+    is_exception_group = BaseExceptionGroup is not None and isinstance(exc_value, BaseExceptionGroup)
+
+    if is_exception_group:
+        (_, exceptions) = exceptions_from_error(
+            exc_type=exc_type,
+            exc_value=exc_value,
+            tb=tb,
+            client_options=client_options,
+            mechanism=mechanism,
+            exception_id=0,
+            parent_id=0,
+        )
+
+    else:
+        exceptions = []
+        for exc_type, exc_value, tb in walk_exception_chain(exc_info):
+            exceptions.append(single_exception_from_error_tuple(exc_type, exc_value, tb, client_options, mechanism))
+
+    exceptions.reverse()
+
+    return exceptions
+
+
+def to_string(value):
+    # type: (str) -> str
+    try:
+        return str(value)
+    except UnicodeDecodeError:
+        return repr(value)[1:-1]
+
+
+def iter_event_stacktraces(event):
+    # type: (Event) -> Iterator[Dict[str, Any]]
+    if "stacktrace" in event:
+        yield event["stacktrace"]
+    if "threads" in event:
+        for thread in event["threads"].get("values") or ():
+            if "stacktrace" in thread:
+                yield thread["stacktrace"]
+    if "exception" in event:
+        for exception in event["exception"].get("values") or ():
+            if "stacktrace" in exception:
+                yield exception["stacktrace"]
+
+
+def iter_event_frames(event):
+    # type: (Event) -> Iterator[Dict[str, Any]]
+    for stacktrace in iter_event_stacktraces(event):
+        for frame in stacktrace.get("frames") or ():
+            yield frame
+
+
+def handle_in_app(event, in_app_exclude=None, in_app_include=None, project_root=None):
+    # type: (Event, Optional[List[str]], Optional[List[str]], Optional[str]) -> Event
+    for stacktrace in iter_event_stacktraces(event):
+        set_in_app_in_frames(
+            stacktrace.get("frames"),
+            in_app_exclude=in_app_exclude,
+            in_app_include=in_app_include,
+            project_root=project_root,
+        )
+
+    return event
+
+
+def set_in_app_in_frames(frames, in_app_exclude, in_app_include, project_root=None):
+    # type: (Any, Optional[List[str]], Optional[List[str]], Optional[str]) -> Optional[Any]
+    if not frames:
+        return None
+
+    for frame in frames:
+        # if frame has already been marked as in_app, skip it
+        current_in_app = frame.get("in_app")
+        if current_in_app is not None:
+            continue
+
+        module = frame.get("module")
+
+        # check if module in frame is in the list of modules to include
+        if _module_in_list(module, in_app_include):
+            frame["in_app"] = True
+            continue
+
+        # check if module in frame is in the list of modules to exclude
+        if _module_in_list(module, in_app_exclude):
+            frame["in_app"] = False
+            continue
+
+        # if frame has no abs_path, skip further checks
+        abs_path = frame.get("abs_path")
+        if abs_path is None:
+            continue
+
+        if _is_external_source(abs_path):
+            frame["in_app"] = False
+            continue
+
+        if _is_in_project_root(abs_path, project_root):
+            frame["in_app"] = True
+            continue
+
+    return frames
+
+
+def exc_info_from_error(error):
+    # type: (Union[BaseException, ExcInfo]) -> ExcInfo
+    if isinstance(error, tuple) and len(error) == 3:
+        exc_type, exc_value, tb = error
+    elif isinstance(error, BaseException):
+        tb = getattr(error, "__traceback__", None)
+        if tb is not None:
+            exc_type = type(error)
+            exc_value = error
+        else:
+            exc_type, exc_value, tb = sys.exc_info()
+            if exc_value is not error:
+                tb = None
+                exc_value = error
+                exc_type = type(error)
+
+    else:
+        raise ValueError("Expected Exception object to report, got %s!" % type(error))
+
+    exc_info = (exc_type, exc_value, tb)
+
+    if TYPE_CHECKING:
+        # This cast is safe because exc_type and exc_value are either both
+        # None or both not None.
+        exc_info = cast(ExcInfo, exc_info)
+
+    return exc_info
+
+
+def event_from_exception(
+    exc_info,  # type: Union[BaseException, ExcInfo]
+    client_options=None,  # type: Optional[Dict[str, Any]]
+    mechanism=None,  # type: Optional[Dict[str, Any]]
+):
+    # type: (...) -> Tuple[Event, Dict[str, Any]]
+    exc_info = exc_info_from_error(exc_info)
+    hint = event_hint_with_exc_info(exc_info)
+    return (
+        {
+            "level": "error",
+            "exception": {"values": exceptions_from_error_tuple(exc_info, client_options, mechanism)},
+        },
+        hint,
+    )
+
+
+def _module_in_list(name, items):
+    # type: (str, Optional[List[str]]) -> bool
+    if name is None:
+        return False
+
+    if not items:
+        return False
+
+    for item in items:
+        if item == name or name.startswith(item + "."):
+            return True
+
+    return False
+
+
+def _is_external_source(abs_path):
+    # type: (str) -> bool
+    # check if frame is in 'site-packages' or 'dist-packages'
+    external_source = re.search(r"[\\/](?:dist|site)-packages[\\/]", abs_path) is not None
+    return external_source
+
+
+def _is_in_project_root(abs_path, project_root):
+    # type: (str, Optional[str]) -> bool
+    if project_root is None:
+        return False
+
+    # check if path is in the project root
+    if abs_path.startswith(project_root):
+        return True
+
+    return False
+
+
+def _truncate_by_bytes(string, max_bytes):
+    # type: (str, int) -> str
+    """
+    Truncate a UTF-8-encodable string to the last full codepoint so that it fits in max_bytes.
+    """
+    truncated = string.encode("utf-8")[: max_bytes - 3].decode("utf-8", errors="ignore")
+
+    return truncated + "..."
+
+
+def _get_size_in_bytes(value):
+    # type: (str) -> Optional[int]
+    try:
+        return len(value.encode("utf-8"))
+    except (UnicodeEncodeError, UnicodeDecodeError):
+        return None
+
+
+def strip_string(value, max_length=None):
+    # type: (str, Optional[int]) -> Union[AnnotatedValue, str]
+    if not value:
+        return value
+
+    if max_length is None:
+        max_length = DEFAULT_MAX_VALUE_LENGTH
+
+    byte_size = _get_size_in_bytes(value)
+    text_size = len(value)
+
+    if byte_size is not None and byte_size > max_length:
+        # truncate to max_length bytes, preserving code points
+        truncated_value = _truncate_by_bytes(value, max_length)
+    elif text_size is not None and text_size > max_length:
+        # fallback to truncating by string length
+        truncated_value = value[: max_length - 3] + "..."
+    else:
+        return value
+
+    return AnnotatedValue(
+        value=truncated_value,
+        metadata={
+            "len": byte_size or text_size,
+            "rem": [["!limit", "x", max_length - 3, max_length]],
+        },
+    )

posthog/feature_flags.py

@@ -13,6 +13,8 @@ __LONG_SCALE__ = float(0xFFFFFFFFFFFFFFF)
 
 log = logging.getLogger("posthog")
 
+NONE_VALUES_ALLOWED_OPERATORS = ["is_not"]
+
 
 class InconclusiveMatchError(Exception):
     pass
@@ -119,6 +121,9 @@ def match_property(property, property_values) -> bool:
 
     override_value = property_values[key]
 
+    if (operator not in NONE_VALUES_ALLOWED_OPERATORS) and override_value is None:
+        return False
+
     if operator in ("exact", "is_not"):
 
         def compute_exact_match(value, override_value):
@@ -175,11 +180,11 @@ def match_property(property, property_values) -> bool:
         else:
             return compare(str(override_value), str(value), operator)
 
-    if operator in ["is_date_before", "is_date_after", "is_relative_date_before", "is_relative_date_after"]:
+    if operator in ["is_date_before", "is_date_after"]:
         try:
-            if operator in ["is_relative_date_before", "is_relative_date_after"]:
-                parsed_date = relative_date_parse_for_feature_flag_matching(str(value))
-            else:
+            parsed_date = relative_date_parse_for_feature_flag_matching(str(value))
+
+            if not parsed_date:
                 parsed_date = parser.parse(str(value))
                 parsed_date = convert_to_datetime_aware(parsed_date)
         except Exception as e:
@@ -190,12 +195,12 @@ def match_property(property, property_values) -> bool:
 
         if isinstance(override_value, datetime.datetime):
             override_date = convert_to_datetime_aware(override_value)
-            if operator in ("is_date_before", "is_relative_date_before"):
+            if operator == "is_date_before":
                 return override_date < parsed_date
             else:
                 return override_date > parsed_date
         elif isinstance(override_value, datetime.date):
-            if operator in ("is_date_before", "is_relative_date_before"):
+            if operator == "is_date_before":
                 return override_value < parsed_date.date()
             else:
                 return override_value > parsed_date.date()
@@ -203,7 +208,7 @@ def match_property(property, property_values) -> bool:
             try:
                 override_date = parser.parse(override_value)
                 override_date = convert_to_datetime_aware(override_date)
-                if operator in ("is_date_before", "is_relative_date_before"):
+                if operator == "is_date_before":
                     return override_date < parsed_date
                 else:
                     return override_date > parsed_date
@@ -302,7 +307,7 @@ def match_property_group(property_group, property_values, cohort_properties) ->
 
 
 def relative_date_parse_for_feature_flag_matching(value: str) -> Optional[datetime.datetime]:
-    regex = r"^(?P<number>[0-9]+)(?P<interval>[a-z])$"
+    regex = r"^-?(?P<number>[0-9]+)(?P<interval>[a-z])$"
     match = re.search(regex, value)
     parsed_dt = datetime.datetime.now(datetime.timezone.utc)
     if match:

posthog/request.py

@@ -13,17 +13,31 @@ from posthog.version import VERSION
 
 _session = requests.sessions.Session()
 
-DEFAULT_HOST = "https://app.posthog.com"
+US_INGESTION_ENDPOINT = "https://us.i.posthog.com"
+EU_INGESTION_ENDPOINT = "https://eu.i.posthog.com"
+DEFAULT_HOST = US_INGESTION_ENDPOINT
 USER_AGENT = "posthog-python/" + VERSION
 
 
+def determine_server_host(host: Optional[str]) -> str:
+    """Determines the server host to use."""
+    host_or_default = host or DEFAULT_HOST
+    trimmed_host = remove_trailing_slash(host_or_default)
+    if trimmed_host in ("https://app.posthog.com", "https://us.posthog.com"):
+        return US_INGESTION_ENDPOINT
+    elif trimmed_host == "https://eu.posthog.com":
+        return EU_INGESTION_ENDPOINT
+    else:
+        return host_or_default
+
+
 def post(
     api_key: str, host: Optional[str] = None, path=None, gzip: bool = False, timeout: int = 15, **kwargs
 ) -> requests.Response:
     """Post the `kwargs` to the API"""
     log = logging.getLogger("posthog")
     body = kwargs
-    body["sentAt"] = datetime.utcnow().replace(tzinfo=tzutc()).isoformat()
+    body["sentAt"] = datetime.now(tz=tzutc()).isoformat()
     url = remove_trailing_slash(host or DEFAULT_HOST) + path
     body["api_key"] = api_key
     data = json.dumps(body, cls=DatetimeSerializer)

posthog/sentry/posthog_integration.py

@@ -43,9 +43,9 @@ class PostHogIntegration(Integration):
                             not not Hub.current.client.dsn and Dsn(Hub.current.client.dsn).project_id
                         )
                         if project_id:
-                            properties[
-                                "$sentry_url"
-                            ] = f"{PostHogIntegration.prefix}{PostHogIntegration.organization}/issues/?project={project_id}&query={event['event_id']}"
+                            properties["$sentry_url"] = (
+                                f"{PostHogIntegration.prefix}{PostHogIntegration.organization}/issues/?project={project_id}&query={event['event_id']}"
+                            )
 
                     posthog.capture(posthog_distinct_id, "$exception", properties)
 

posthog/test/ai/anthropic/test_anthropic.py

@@ -0,0 +1,327 @@
+import os
+import time
+from unittest.mock import patch
+
+import pytest
+from anthropic.types import Message, Usage
+
+from posthog.ai.anthropic import Anthropic, AsyncAnthropic
+
+ANTHROPIC_API_KEY = os.getenv("ANTHROPIC_API_KEY")
+
+
+@pytest.fixture
+def mock_client():
+    with patch("posthog.client.Client") as mock_client:
+        mock_client.privacy_mode = False
+        yield mock_client
+
+
+@pytest.fixture
+def mock_anthropic_response():
+    return Message(
+        id="msg_123",
+        type="message",
+        role="assistant",
+        content=[{"type": "text", "text": "Test response"}],
+        model="claude-3-opus-20240229",
+        usage=Usage(
+            input_tokens=20,
+            output_tokens=10,
+        ),
+        stop_reason="end_turn",
+        stop_sequence=None,
+    )
+
+
+@pytest.fixture
+def mock_anthropic_stream():
+    class MockStreamEvent:
+        def __init__(self, content, usage=None):
+            self.content = content
+            self.usage = usage
+
+    def stream_generator():
+        yield MockStreamEvent("A")
+        yield MockStreamEvent("B")
+        yield MockStreamEvent(
+            "C",
+            usage=Usage(
+                input_tokens=20,
+                output_tokens=10,
+            ),
+        )
+
+    return stream_generator()
+
+
+def test_basic_completion(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)
+        response = client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_properties={"foo": "bar"},
+        )
+
+        assert response == mock_anthropic_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"] == "anthropic"
+        assert props["$ai_model"] == "claude-3-opus-20240229"
+        assert props["$ai_input"] == [{"role": "user", "content": "Hello"}]
+        assert props["$ai_output_choices"] == [{"role": "assistant", "content": "Test response"}]
+        assert props["$ai_input_tokens"] == 20
+        assert props["$ai_output_tokens"] == 10
+        assert props["$ai_http_status"] == 200
+        assert props["foo"] == "bar"
+        assert isinstance(props["$ai_latency"], float)
+
+
+def test_streaming(mock_client, mock_anthropic_stream):
+    with patch("anthropic.resources.Messages.create", return_value=mock_anthropic_stream):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        response = client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            stream=True,
+            posthog_distinct_id="test-id",
+            posthog_properties={"foo": "bar"},
+        )
+
+        # Consume the stream
+        chunks = list(response)
+        assert len(chunks) == 3
+        assert chunks[0].content == "A"
+        assert chunks[1].content == "B"
+        assert chunks[2].content == "C"
+
+        # Wait a bit to ensure the capture is called
+        time.sleep(0.1)
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+
+        assert call_args["distinct_id"] == "test-id"
+        assert call_args["event"] == "$ai_generation"
+        assert props["$ai_provider"] == "anthropic"
+        assert props["$ai_model"] == "claude-3-opus-20240229"
+        assert props["$ai_input"] == [{"role": "user", "content": "Hello"}]
+        assert props["$ai_output_choices"] == [{"role": "assistant", "content": "ABC"}]
+        assert props["$ai_input_tokens"] == 20
+        assert props["$ai_output_tokens"] == 10
+        assert isinstance(props["$ai_latency"], float)
+        assert props["foo"] == "bar"
+
+
+def test_streaming_with_stream_endpoint(mock_client, mock_anthropic_stream):
+    with patch("anthropic.resources.Messages.create", return_value=mock_anthropic_stream):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        response = client.messages.stream(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_properties={"foo": "bar"},
+        )
+
+        # Consume the stream
+        chunks = list(response)
+        assert len(chunks) == 3
+        assert chunks[0].content == "A"
+        assert chunks[1].content == "B"
+        assert chunks[2].content == "C"
+
+        # Wait a bit to ensure the capture is called
+        time.sleep(0.1)
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+
+        assert call_args["distinct_id"] == "test-id"
+        assert call_args["event"] == "$ai_generation"
+        assert props["$ai_provider"] == "anthropic"
+        assert props["$ai_model"] == "claude-3-opus-20240229"
+        assert props["$ai_input"] == [{"role": "user", "content": "Hello"}]
+        assert props["$ai_output_choices"] == [{"role": "assistant", "content": "ABC"}]
+        assert props["$ai_input_tokens"] == 20
+        assert props["$ai_output_tokens"] == 10
+        assert isinstance(props["$ai_latency"], float)
+        assert props["foo"] == "bar"
+
+
+def test_groups(mock_client, mock_anthropic_response):
+    with patch("anthropic.resources.Messages.create", return_value=mock_anthropic_response):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        response = client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_groups={"company": "test_company"},
+        )
+
+        assert response == mock_anthropic_response
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        assert call_args["groups"] == {"company": "test_company"}
+
+
+def test_privacy_mode_local(mock_client, mock_anthropic_response):
+    with patch("anthropic.resources.Messages.create", return_value=mock_anthropic_response):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        response = client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_privacy_mode=True,
+        )
+
+        assert response == mock_anthropic_response
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+        assert props["$ai_input"] is None
+        assert props["$ai_output_choices"] is None
+
+
+def test_privacy_mode_global(mock_client, mock_anthropic_response):
+    with patch("anthropic.resources.Messages.create", return_value=mock_anthropic_response):
+        mock_client.privacy_mode = True
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        response = client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_privacy_mode=False,
+        )
+
+        assert response == mock_anthropic_response
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+        assert props["$ai_input"] is None
+        assert props["$ai_output_choices"] is None
+
+
+@pytest.mark.skipif(not ANTHROPIC_API_KEY, reason="ANTHROPIC_API_KEY is not set")
+def test_basic_integration(mock_client):
+    client = Anthropic(posthog_client=mock_client)
+    client.messages.create(
+        model="claude-3-opus-20240229",
+        messages=[{"role": "user", "content": "Foo"}],
+        max_tokens=1,
+        temperature=0,
+        posthog_distinct_id="test-id",
+        posthog_properties={"foo": "bar"},
+        system="You must always answer with 'Bar'.",
+    )
+
+    assert mock_client.capture.call_count == 1
+
+    call_args = mock_client.capture.call_args[1]
+    props = call_args["properties"]
+    assert call_args["distinct_id"] == "test-id"
+    assert call_args["event"] == "$ai_generation"
+    assert props["$ai_provider"] == "anthropic"
+    assert props["$ai_model"] == "claude-3-opus-20240229"
+    assert props["$ai_input"] == [
+        {"role": "system", "content": "You must always answer with 'Bar'."},
+        {"role": "user", "content": "Foo"},
+    ]
+    assert props["$ai_output_choices"][0]["role"] == "assistant"
+    assert props["$ai_output_choices"][0]["content"] == "Bar"
+    assert props["$ai_input_tokens"] == 18
+    assert props["$ai_output_tokens"] == 1
+    assert props["$ai_http_status"] == 200
+    assert props["foo"] == "bar"
+    assert isinstance(props["$ai_latency"], float)
+
+
+@pytest.mark.skipif(not ANTHROPIC_API_KEY, reason="ANTHROPIC_API_KEY is not set")
+async def test_basic_async_integration(mock_client):
+    client = AsyncAnthropic(posthog_client=mock_client)
+    await client.messages.create(
+        model="claude-3-opus-20240229",
+        messages=[{"role": "user", "content": "You must always answer with 'Bar'."}],
+        max_tokens=1,
+        temperature=0,
+        posthog_distinct_id="test-id",
+        posthog_properties={"foo": "bar"},
+    )
+
+    assert mock_client.capture.call_count == 1
+
+    call_args = mock_client.capture.call_args[1]
+    props = call_args["properties"]
+
+    assert call_args["distinct_id"] == "test-id"
+    assert call_args["event"] == "$ai_generation"
+    assert props["$ai_provider"] == "anthropic"
+    assert props["$ai_model"] == "claude-3-opus-20240229"
+    assert props["$ai_input"] == [{"role": "user", "content": "You must always answer with 'Bar'."}]
+    assert props["$ai_output_choices"][0]["role"] == "assistant"
+    assert props["$ai_input_tokens"] == 16
+    assert props["$ai_output_tokens"] == 1
+    assert props["$ai_http_status"] == 200
+    assert props["foo"] == "bar"
+    assert isinstance(props["$ai_latency"], float)
+
+
+def test_streaming_system_prompt(mock_client, mock_anthropic_stream):
+    with patch("anthropic.resources.Messages.create", return_value=mock_anthropic_stream):
+        client = Anthropic(api_key="test-key", posthog_client=mock_client)
+        response = client.messages.create(
+            model="claude-3-opus-20240229",
+            system="Foo",
+            messages=[{"role": "user", "content": "Bar"}],
+            stream=True,
+        )
+
+        # Consume the stream
+        list(response)
+
+        # Wait a bit to ensure the capture is called
+        time.sleep(0.1)
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+
+        assert props["$ai_input"] == [{"role": "system", "content": "Foo"}, {"role": "user", "content": "Bar"}]
+
+
+@pytest.mark.skipif(not ANTHROPIC_API_KEY, reason="ANTHROPIC_API_KEY is not set")
+async def test_async_streaming_system_prompt(mock_client, mock_anthropic_stream):
+    client = AsyncAnthropic(posthog_client=mock_client)
+    response = await client.messages.create(
+        model="claude-3-opus-20240229",
+        system="You must always answer with 'Bar'.",
+        messages=[{"role": "user", "content": "Foo"}],
+        stream=True,
+        max_tokens=1,
+    )
+
+    # Consume the stream
+    [c async for c in response]
+
+    # Wait a bit to ensure the capture is called
+    time.sleep(0.1)
+    assert mock_client.capture.call_count == 1
+
+    call_args = mock_client.capture.call_args[1]
+    props = call_args["properties"]
+
+    assert props["$ai_input"] == [
+        {"role": "system", "content": "You must always answer with 'Bar'."},
+        {"role": "user", "content": "Foo"},
+    ]

posthog/test/ai/langchain/__init__.py

@@ -0,0 +1,4 @@
+import pytest
+
+pytest.importorskip("langchain")
+pytest.importorskip("langchain_community")

posthog/test/ai/langchain/test_callbacks.py

@@ -0,0 +1,773 @@
+import math
+import os
+import time
+import uuid
+from unittest.mock import patch
+
+import pytest
+from langchain_anthropic.chat_models import ChatAnthropic
+from langchain_community.chat_models.fake import FakeMessagesListChatModel
+from langchain_community.llms.fake import FakeListLLM, FakeStreamingListLLM
+from langchain_core.messages import AIMessage
+from langchain_core.prompts import ChatPromptTemplate
+from langchain_core.runnables import RunnableLambda
+from langchain_openai.chat_models import ChatOpenAI
+
+from posthog.ai.langchain import CallbackHandler
+
+OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
+ANTHROPIC_API_KEY = os.getenv("ANTHROPIC_API_KEY")
+
+
+@pytest.fixture(scope="function")
+def mock_client():
+    with patch("posthog.client.Client") as mock_client:
+        mock_client.privacy_mode = False
+        yield mock_client
+
+
+def test_parent_capture(mock_client):
+    callbacks = CallbackHandler(mock_client)
+    parent_run_id = uuid.uuid4()
+    run_id = uuid.uuid4()
+    callbacks._set_parent_of_run(run_id, parent_run_id)
+    assert callbacks._parent_tree == {run_id: parent_run_id}
+    callbacks._pop_parent_of_run(run_id)
+    assert callbacks._parent_tree == {}
+    callbacks._pop_parent_of_run(parent_run_id)  # should not raise
+
+
+def test_find_root_run(mock_client):
+    callbacks = CallbackHandler(mock_client)
+    root_run_id = uuid.uuid4()
+    parent_run_id = uuid.uuid4()
+    run_id = uuid.uuid4()
+    callbacks._set_parent_of_run(run_id, parent_run_id)
+    callbacks._set_parent_of_run(parent_run_id, root_run_id)
+    assert callbacks._find_root_run(run_id) == root_run_id
+    new_run_id = uuid.uuid4()
+    assert callbacks._find_root_run(new_run_id) == new_run_id
+
+
+def test_trace_id_generation(mock_client):
+    callbacks = CallbackHandler(mock_client)
+    run_id = uuid.uuid4()
+    with patch("uuid.uuid4", return_value=run_id):
+        assert callbacks._get_trace_id(run_id) == run_id
+    run_id = uuid.uuid4()
+    callbacks = CallbackHandler(mock_client, trace_id=run_id)
+    assert callbacks._get_trace_id(uuid.uuid4()) == run_id
+
+
+def test_metadata_capture(mock_client):
+    callbacks = CallbackHandler(mock_client)
+    run_id = uuid.uuid4()
+    with patch("time.time", return_value=1234567890):
+        callbacks._set_run_metadata(
+            {"kwargs": {"openai_api_base": "https://us.posthog.com"}},
+            run_id,
+            messages=[{"role": "user", "content": "Who won the world series in 2020?"}],
+            invocation_params={"temperature": 0.5},
+            metadata={"ls_model_name": "hog-mini", "ls_provider": "posthog"},
+        )
+    expected = {
+        "model": "hog-mini",
+        "messages": [{"role": "user", "content": "Who won the world series in 2020?"}],
+        "start_time": 1234567890,
+        "model_params": {"temperature": 0.5},
+        "provider": "posthog",
+        "base_url": "https://us.posthog.com",
+    }
+    assert callbacks._runs[run_id] == expected
+    with patch("time.time", return_value=1234567891):
+        run = callbacks._pop_run_metadata(run_id)
+    assert run == {**expected, "end_time": 1234567891}
+    assert callbacks._runs == {}
+    callbacks._pop_run_metadata(uuid.uuid4())  # should not raise
+
+
+@pytest.mark.parametrize("stream", [True, False])
+def test_basic_chat_chain(mock_client, stream):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", "You are a helpful assistant."),
+            ("user", "Who won the world series in 2020?"),
+        ]
+    )
+    model = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="The Los Angeles Dodgers won the World Series in 2020.",
+                usage_metadata={"input_tokens": 10, "output_tokens": 10, "total_tokens": 20},
+            )
+        ]
+    )
+    callbacks = [CallbackHandler(mock_client)]
+    chain = prompt | model
+    if stream:
+        result = [m for m in chain.stream({}, config={"callbacks": callbacks})][0]
+    else:
+        result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "The Los Angeles Dodgers won the World Series in 2020."
+    assert mock_client.capture.call_count == 1
+    args = mock_client.capture.call_args[1]
+    props = args["properties"]
+
+    assert args["event"] == "$ai_generation"
+    assert "distinct_id" in args
+    assert "$ai_model" in props
+    assert "$ai_provider" in props
+    assert props["$ai_input"] == [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": "Who won the world series in 2020?"},
+    ]
+    assert props["$ai_output_choices"] == [
+        {"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."}
+    ]
+    assert props["$ai_input_tokens"] == 10
+    assert props["$ai_output_tokens"] == 10
+    assert props["$ai_http_status"] == 200
+    assert props["$ai_trace_id"] is not None
+    assert isinstance(props["$ai_latency"], float)
+
+
+@pytest.mark.parametrize("stream", [True, False])
+async def test_async_basic_chat_chain(mock_client, stream):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", "You are a helpful assistant."),
+            ("user", "Who won the world series in 2020?"),
+        ]
+    )
+    model = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="The Los Angeles Dodgers won the World Series in 2020.",
+                usage_metadata={"input_tokens": 10, "output_tokens": 10, "total_tokens": 20},
+            )
+        ]
+    )
+    callbacks = [CallbackHandler(mock_client)]
+    chain = prompt | model
+    if stream:
+        result = [m async for m in chain.astream({}, config={"callbacks": callbacks})][0]
+    else:
+        result = await chain.ainvoke({}, config={"callbacks": callbacks})
+    assert result.content == "The Los Angeles Dodgers won the World Series in 2020."
+    assert mock_client.capture.call_count == 1
+
+    args = mock_client.capture.call_args[1]
+    props = args["properties"]
+    assert args["event"] == "$ai_generation"
+    assert "distinct_id" in args
+    assert "$ai_model" in props
+    assert "$ai_provider" in props
+    assert props["$ai_input"] == [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": "Who won the world series in 2020?"},
+    ]
+    assert props["$ai_output_choices"] == [
+        {"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."}
+    ]
+    assert props["$ai_input_tokens"] == 10
+    assert props["$ai_output_tokens"] == 10
+    assert props["$ai_http_status"] == 200
+    assert props["$ai_trace_id"] is not None
+    assert isinstance(props["$ai_latency"], float)
+
+
+@pytest.mark.parametrize(
+    "Model,stream",
+    [(FakeListLLM, True), (FakeListLLM, False), (FakeStreamingListLLM, True), (FakeStreamingListLLM, False)],
+)
+def test_basic_llm_chain(mock_client, Model, stream):
+    model = Model(responses=["The Los Angeles Dodgers won the World Series in 2020."])
+    callbacks: list[CallbackHandler] = [CallbackHandler(mock_client)]
+
+    if stream:
+        result = "".join(
+            [m for m in model.stream("Who won the world series in 2020?", config={"callbacks": callbacks})]
+        )
+    else:
+        result = model.invoke("Who won the world series in 2020?", config={"callbacks": callbacks})
+    assert result == "The Los Angeles Dodgers won the World Series in 2020."
+
+    assert mock_client.capture.call_count == 1
+    args = mock_client.capture.call_args[1]
+    props = args["properties"]
+
+    assert args["event"] == "$ai_generation"
+    assert "distinct_id" in args
+    assert "$ai_model" in props
+    assert "$ai_provider" in props
+    assert props["$ai_input"] == ["Who won the world series in 2020?"]
+    assert props["$ai_output_choices"] == ["The Los Angeles Dodgers won the World Series in 2020."]
+    assert props["$ai_http_status"] == 200
+    assert props["$ai_trace_id"] is not None
+    assert isinstance(props["$ai_latency"], float)
+
+
+@pytest.mark.parametrize(
+    "Model,stream",
+    [(FakeListLLM, True), (FakeListLLM, False), (FakeStreamingListLLM, True), (FakeStreamingListLLM, False)],
+)
+async def test_async_basic_llm_chain(mock_client, Model, stream):
+    model = Model(responses=["The Los Angeles Dodgers won the World Series in 2020."])
+    callbacks: list[CallbackHandler] = [CallbackHandler(mock_client)]
+
+    if stream:
+        result = "".join(
+            [m async for m in model.astream("Who won the world series in 2020?", config={"callbacks": callbacks})]
+        )
+    else:
+        result = await model.ainvoke("Who won the world series in 2020?", config={"callbacks": callbacks})
+    assert result == "The Los Angeles Dodgers won the World Series in 2020."
+
+    assert mock_client.capture.call_count == 1
+    args = mock_client.capture.call_args[1]
+    props = args["properties"]
+
+    assert args["event"] == "$ai_generation"
+    assert "distinct_id" in args
+    assert "$ai_model" in props
+    assert "$ai_provider" in props
+    assert props["$ai_input"] == ["Who won the world series in 2020?"]
+    assert props["$ai_output_choices"] == ["The Los Angeles Dodgers won the World Series in 2020."]
+    assert props["$ai_http_status"] == 200
+    assert props["$ai_trace_id"] is not None
+    assert isinstance(props["$ai_latency"], float)
+
+
+def test_trace_id_for_multiple_chains(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("user", "Foo"),
+        ]
+    )
+    model = FakeMessagesListChatModel(responses=[AIMessage(content="Bar")])
+    callbacks = [CallbackHandler(mock_client)]
+    chain = prompt | model | RunnableLambda(lambda x: [x]) | model
+    result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "Bar"
+    assert mock_client.capture.call_count == 2
+
+    first_call_args = mock_client.capture.call_args_list[0][1]
+    first_call_props = first_call_args["properties"]
+    assert first_call_args["event"] == "$ai_generation"
+    assert "distinct_id" in first_call_args
+    assert "$ai_model" in first_call_props
+    assert "$ai_provider" in first_call_props
+    assert first_call_props["$ai_input"] == [{"role": "user", "content": "Foo"}]
+    assert first_call_props["$ai_output_choices"] == [{"role": "assistant", "content": "Bar"}]
+    assert first_call_props["$ai_http_status"] == 200
+    assert first_call_props["$ai_trace_id"] is not None
+    assert isinstance(first_call_props["$ai_latency"], float)
+
+    second_call_args = mock_client.capture.call_args_list[1][1]
+    second_call_props = second_call_args["properties"]
+    assert second_call_args["event"] == "$ai_generation"
+    assert "distinct_id" in second_call_args
+    assert "$ai_model" in second_call_props
+    assert "$ai_provider" in second_call_props
+    assert second_call_props["$ai_input"] == [{"role": "assistant", "content": "Bar"}]
+    assert second_call_props["$ai_output_choices"] == [{"role": "assistant", "content": "Bar"}]
+    assert second_call_props["$ai_http_status"] == 200
+    assert second_call_props["$ai_trace_id"] is not None
+    assert isinstance(second_call_props["$ai_latency"], float)
+
+    # Check that the trace_id is the same as the first call
+    assert first_call_props["$ai_trace_id"] == second_call_props["$ai_trace_id"]
+
+
+def test_personless_mode(mock_client):
+    prompt = ChatPromptTemplate.from_messages([("user", "Foo")])
+    chain = prompt | FakeMessagesListChatModel(responses=[AIMessage(content="Bar")])
+    chain.invoke({}, config={"callbacks": [CallbackHandler(mock_client)]})
+    assert mock_client.capture.call_count == 1
+    args = mock_client.capture.call_args_list[0][1]
+    assert args["properties"]["$process_person_profile"] is False
+
+    id = uuid.uuid4()
+    chain.invoke({}, config={"callbacks": [CallbackHandler(mock_client, distinct_id=id)]})
+    assert mock_client.capture.call_count == 2
+    args = mock_client.capture.call_args_list[1][1]
+    assert "$process_person_profile" not in args["properties"]
+    assert args["distinct_id"] == id
+
+
+def test_personless_mode_exception(mock_client):
+    prompt = ChatPromptTemplate.from_messages([("user", "Foo")])
+    chain = prompt | ChatOpenAI(api_key="test", model="gpt-4o-mini")
+    callbacks = CallbackHandler(mock_client)
+    with pytest.raises(Exception):
+        chain.invoke({}, config={"callbacks": [callbacks]})
+    assert mock_client.capture.call_count == 1
+    args = mock_client.capture.call_args_list[0][1]
+    assert args["properties"]["$process_person_profile"] is False
+
+    id = uuid.uuid4()
+    with pytest.raises(Exception):
+        chain.invoke({}, config={"callbacks": [CallbackHandler(mock_client, distinct_id=id)]})
+    assert mock_client.capture.call_count == 2
+    args = mock_client.capture.call_args_list[1][1]
+    assert "$process_person_profile" not in args["properties"]
+    assert args["distinct_id"] == id
+
+
+def test_metadata(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("user", "Foo"),
+        ]
+    )
+    model = FakeMessagesListChatModel(responses=[AIMessage(content="Bar")])
+    callbacks = [
+        CallbackHandler(mock_client, trace_id="test-trace-id", distinct_id="test_id", properties={"foo": "bar"})
+    ]
+    chain = prompt | model
+    result = chain.invoke({}, config={"callbacks": callbacks})
+
+    assert result.content == "Bar"
+    assert mock_client.capture.call_count == 1
+
+    first_call_args = mock_client.capture.call_args[1]
+    assert first_call_args["distinct_id"] == "test_id"
+
+    first_call_props = first_call_args["properties"]
+    assert first_call_args["event"] == "$ai_generation"
+    assert first_call_props["$ai_trace_id"] == "test-trace-id"
+    assert first_call_props["foo"] == "bar"
+    assert first_call_props["$ai_input"] == [{"role": "user", "content": "Foo"}]
+    assert first_call_props["$ai_output_choices"] == [{"role": "assistant", "content": "Bar"}]
+    assert first_call_props["$ai_http_status"] == 200
+    assert isinstance(first_call_props["$ai_latency"], float)
+
+
+def test_callbacks_logic(mock_client):
+    prompt = ChatPromptTemplate.from_messages([("user", "Foo")])
+    model = FakeMessagesListChatModel(responses=[AIMessage(content="Bar")])
+    callbacks = CallbackHandler(mock_client, trace_id="test-trace-id", distinct_id="test_id", properties={"foo": "bar"})
+    chain = prompt | model
+
+    chain.invoke({}, config={"callbacks": [callbacks]})
+    assert callbacks._runs == {}
+    assert callbacks._parent_tree == {}
+
+    def assert_intermediary_run(m):
+        assert callbacks._runs == {}
+        assert len(callbacks._parent_tree.items()) == 1
+        return [m]
+
+    (chain | RunnableLambda(assert_intermediary_run) | model).invoke({}, config={"callbacks": [callbacks]})
+    assert callbacks._runs == {}
+    assert callbacks._parent_tree == {}
+
+
+def test_exception_in_chain(mock_client):
+    def runnable(_):
+        raise ValueError("test")
+
+    callbacks = CallbackHandler(mock_client)
+    with pytest.raises(ValueError):
+        RunnableLambda(runnable).invoke({}, config={"callbacks": [callbacks]})
+
+    assert callbacks._runs == {}
+    assert callbacks._parent_tree == {}
+    assert mock_client.capture.call_count == 0
+
+
+def test_openai_error(mock_client):
+    prompt = ChatPromptTemplate.from_messages([("user", "Foo")])
+    chain = prompt | ChatOpenAI(api_key="test", model="gpt-4o-mini")
+    callbacks = CallbackHandler(mock_client)
+
+    # 401
+    with pytest.raises(Exception):
+        chain.invoke({}, config={"callbacks": [callbacks]})
+
+    assert callbacks._runs == {}
+    assert callbacks._parent_tree == {}
+    assert mock_client.capture.call_count == 1
+    args = mock_client.capture.call_args[1]
+    props = args["properties"]
+    assert props["$ai_http_status"] == 401
+    assert props["$ai_input"] == [{"role": "user", "content": "Foo"}]
+    assert "$ai_output_choices" not in props
+
+
+@pytest.mark.skipif(not OPENAI_API_KEY, reason="OpenAI API key not set")
+def test_openai_chain(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    chain = prompt | ChatOpenAI(
+        api_key=OPENAI_API_KEY,
+        model="gpt-4o-mini",
+        temperature=0,
+        max_tokens=1,
+    )
+    callbacks = CallbackHandler(mock_client, trace_id="test-trace-id", distinct_id="test_id", properties={"foo": "bar"})
+    start_time = time.time()
+    result = chain.invoke({}, config={"callbacks": [callbacks]})
+    approximate_latency = math.floor(time.time() - start_time)
+
+    assert result.content == "Bar"
+    assert mock_client.capture.call_count == 1
+
+    first_call_args = mock_client.capture.call_args[1]
+    first_call_props = first_call_args["properties"]
+    assert first_call_args["event"] == "$ai_generation"
+    assert first_call_props["$ai_trace_id"] == "test-trace-id"
+    assert first_call_props["$ai_provider"] == "openai"
+    assert first_call_props["$ai_model"] == "gpt-4o-mini"
+    assert first_call_props["foo"] == "bar"
+
+    # langchain-openai for langchain v3
+    if "max_completion_tokens" in first_call_props["$ai_model_parameters"]:
+        assert first_call_props["$ai_model_parameters"] == {
+            "temperature": 0.0,
+            "max_completion_tokens": 1,
+            "stream": False,
+        }
+    else:
+        assert first_call_props["$ai_model_parameters"] == {
+            "temperature": 0.0,
+            "max_tokens": 1,
+            "n": 1,
+            "stream": False,
+        }
+    assert first_call_props["$ai_input"] == [
+        {"role": "system", "content": 'You must always answer with "Bar".'},
+        {"role": "user", "content": "Foo"},
+    ]
+    assert first_call_props["$ai_output_choices"] == [
+        {
+            "role": "assistant",
+            "content": "Bar",
+            "additional_kwargs": {"refusal": None},
+        }
+    ]
+    assert first_call_props["$ai_http_status"] == 200
+    assert isinstance(first_call_props["$ai_latency"], float)
+    assert min(approximate_latency - 1, 0) <= math.floor(first_call_props["$ai_latency"]) <= approximate_latency
+    assert first_call_props["$ai_input_tokens"] == 20
+    assert first_call_props["$ai_output_tokens"] == 1
+
+
+@pytest.mark.skipif(not OPENAI_API_KEY, reason="OpenAI API key not set")
+def test_openai_captures_multiple_generations(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    chain = prompt | ChatOpenAI(
+        api_key=OPENAI_API_KEY,
+        model="gpt-4o-mini",
+        temperature=0,
+        max_tokens=1,
+        n=2,
+    )
+    callbacks = CallbackHandler(mock_client)
+    result = chain.invoke({}, config={"callbacks": [callbacks]})
+
+    assert result.content == "Bar"
+    assert mock_client.capture.call_count == 1
+
+    first_call_args = mock_client.capture.call_args[1]
+    first_call_props = first_call_args["properties"]
+    assert first_call_props["$ai_input"] == [
+        {"role": "system", "content": 'You must always answer with "Bar".'},
+        {"role": "user", "content": "Foo"},
+    ]
+    assert first_call_props["$ai_output_choices"] == [
+        {
+            "role": "assistant",
+            "content": "Bar",
+            "additional_kwargs": {"refusal": None},
+        },
+        {
+            "role": "assistant",
+            "content": "Bar",
+        },
+    ]
+
+    # langchain-openai for langchain v3
+    if "max_completion_tokens" in first_call_props["$ai_model_parameters"]:
+        assert first_call_props["$ai_model_parameters"] == {
+            "temperature": 0.0,
+            "max_completion_tokens": 1,
+            "stream": False,
+            "n": 2,
+        }
+    else:
+        assert first_call_props["$ai_model_parameters"] == {
+            "temperature": 0.0,
+            "max_tokens": 1,
+            "stream": False,
+            "n": 2,
+        }
+    assert first_call_props["$ai_http_status"] == 200
+
+
+@pytest.mark.skipif(not OPENAI_API_KEY, reason="OpenAI API key not set")
+def test_openai_streaming(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    chain = prompt | ChatOpenAI(
+        api_key=OPENAI_API_KEY, model="gpt-4o-mini", temperature=0, max_tokens=1, stream=True, stream_usage=True
+    )
+    callbacks = CallbackHandler(mock_client)
+    result = [m for m in chain.stream({}, config={"callbacks": [callbacks]})]
+    result = sum(result[1:], result[0])
+
+    assert result.content == "Bar"
+    assert mock_client.capture.call_count == 1
+
+    first_call_args = mock_client.capture.call_args[1]
+    first_call_props = first_call_args["properties"]
+
+    assert first_call_props["$ai_model_parameters"]["stream"]
+    assert first_call_props["$ai_input"] == [
+        {"role": "system", "content": 'You must always answer with "Bar".'},
+        {"role": "user", "content": "Foo"},
+    ]
+    assert first_call_props["$ai_output_choices"] == [{"role": "assistant", "content": "Bar"}]
+    assert first_call_props["$ai_http_status"] == 200
+    assert first_call_props["$ai_input_tokens"] == 20
+    assert first_call_props["$ai_output_tokens"] == 1
+
+
+@pytest.mark.skipif(not OPENAI_API_KEY, reason="OpenAI API key not set")
+async def test_async_openai_streaming(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    chain = prompt | ChatOpenAI(
+        api_key=OPENAI_API_KEY, model="gpt-4o-mini", temperature=0, max_tokens=1, stream=True, stream_usage=True
+    )
+    callbacks = CallbackHandler(mock_client)
+    result = [m async for m in chain.astream({}, config={"callbacks": [callbacks]})]
+    result = sum(result[1:], result[0])
+
+    assert result.content == "Bar"
+    assert mock_client.capture.call_count == 1
+
+    first_call_args = mock_client.capture.call_args[1]
+    first_call_props = first_call_args["properties"]
+
+    assert first_call_props["$ai_model_parameters"]["stream"]
+    assert first_call_props["$ai_input"] == [
+        {"role": "system", "content": 'You must always answer with "Bar".'},
+        {"role": "user", "content": "Foo"},
+    ]
+    assert first_call_props["$ai_output_choices"] == [{"role": "assistant", "content": "Bar"}]
+    assert first_call_props["$ai_http_status"] == 200
+    assert first_call_props["$ai_input_tokens"] == 20
+    assert first_call_props["$ai_output_tokens"] == 1
+
+
+def test_base_url_retrieval(mock_client):
+    prompt = ChatPromptTemplate.from_messages([("user", "Foo")])
+    chain = prompt | ChatOpenAI(
+        api_key="test",
+        model="posthog-mini",
+        base_url="https://test.posthog.com",
+    )
+    callbacks = CallbackHandler(mock_client)
+    with pytest.raises(Exception):
+        chain.invoke({}, config={"callbacks": [callbacks]})
+
+    assert mock_client.capture.call_count == 1
+    call = mock_client.capture.call_args[1]
+    assert call["properties"]["$ai_base_url"] == "https://test.posthog.com"
+
+
+def test_groups(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    model = FakeMessagesListChatModel(responses=[AIMessage(content="Bar")])
+    chain = prompt | model
+    callbacks = CallbackHandler(mock_client, groups={"company": "test_company"})
+    chain.invoke({}, config={"callbacks": [callbacks]})
+
+    assert mock_client.capture.call_count == 1
+    call = mock_client.capture.call_args[1]
+    assert call["groups"] == {"company": "test_company"}
+
+
+def test_privacy_mode_local(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    model = FakeMessagesListChatModel(responses=[AIMessage(content="Bar")])
+    chain = prompt | model
+    callbacks = CallbackHandler(mock_client, privacy_mode=True)
+    chain.invoke({}, config={"callbacks": [callbacks]})
+
+    assert mock_client.capture.call_count == 1
+    call = mock_client.capture.call_args[1]
+    assert call["properties"]["$ai_input"] is None
+    assert call["properties"]["$ai_output_choices"] is None
+
+
+def test_privacy_mode_global(mock_client):
+    mock_client.privacy_mode = True
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    model = FakeMessagesListChatModel(responses=[AIMessage(content="Bar")])
+    chain = prompt | model
+    callbacks = CallbackHandler(mock_client)
+    chain.invoke({}, config={"callbacks": [callbacks]})
+
+    assert mock_client.capture.call_count == 1
+    call = mock_client.capture.call_args[1]
+    assert call["properties"]["$ai_input"] is None
+    assert call["properties"]["$ai_output_choices"] is None
+
+
+@pytest.mark.skipif(not ANTHROPIC_API_KEY, reason="ANTHROPIC_API_KEY is not set")
+def test_anthropic_chain(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    chain = prompt | ChatAnthropic(
+        api_key=ANTHROPIC_API_KEY,
+        model="claude-3-opus-20240229",
+        temperature=0,
+        max_tokens=1,
+    )
+    callbacks = CallbackHandler(mock_client, trace_id="test-trace-id", distinct_id="test_id", properties={"foo": "bar"})
+    start_time = time.time()
+    result = chain.invoke({}, config={"callbacks": [callbacks]})
+    approximate_latency = math.floor(time.time() - start_time)
+
+    assert result.content == "Bar"
+    assert mock_client.capture.call_count == 1
+
+    first_call_args = mock_client.capture.call_args[1]
+    first_call_props = first_call_args["properties"]
+    assert first_call_args["event"] == "$ai_generation"
+    assert first_call_props["$ai_trace_id"] == "test-trace-id"
+    assert first_call_props["$ai_provider"] == "anthropic"
+    assert first_call_props["$ai_model"] == "claude-3-opus-20240229"
+    assert first_call_props["foo"] == "bar"
+
+    assert first_call_props["$ai_model_parameters"] == {
+        "temperature": 0.0,
+        "max_tokens": 1,
+        "streaming": False,
+    }
+    assert first_call_props["$ai_input"] == [
+        {"role": "system", "content": 'You must always answer with "Bar".'},
+        {"role": "user", "content": "Foo"},
+    ]
+    assert first_call_props["$ai_output_choices"] == [{"role": "assistant", "content": "Bar"}]
+    assert first_call_props["$ai_http_status"] == 200
+    assert isinstance(first_call_props["$ai_latency"], float)
+    assert min(approximate_latency - 1, 0) <= math.floor(first_call_props["$ai_latency"]) <= approximate_latency
+    assert first_call_props["$ai_input_tokens"] == 17
+    assert first_call_props["$ai_output_tokens"] == 1
+
+
+@pytest.mark.skipif(not ANTHROPIC_API_KEY, reason="ANTHROPIC_API_KEY is not set")
+async def test_async_anthropic_streaming(mock_client):
+    prompt = ChatPromptTemplate.from_messages(
+        [
+            ("system", 'You must always answer with "Bar".'),
+            ("user", "Foo"),
+        ]
+    )
+    chain = prompt | ChatAnthropic(
+        api_key=ANTHROPIC_API_KEY,
+        model="claude-3-opus-20240229",
+        temperature=0,
+        max_tokens=1,
+        streaming=True,
+        stream_usage=True,
+    )
+    callbacks = CallbackHandler(mock_client)
+    result = [m async for m in chain.astream({}, config={"callbacks": [callbacks]})]
+    result = sum(result[1:], result[0])
+
+    assert result.content == "Bar"
+    assert mock_client.capture.call_count == 1
+
+    first_call_args = mock_client.capture.call_args[1]
+    first_call_props = first_call_args["properties"]
+    assert first_call_props["$ai_model_parameters"]["streaming"]
+    assert first_call_props["$ai_input"] == [
+        {"role": "system", "content": 'You must always answer with "Bar".'},
+        {"role": "user", "content": "Foo"},
+    ]
+    assert first_call_props["$ai_output_choices"] == [{"role": "assistant", "content": "Bar"}]
+    assert first_call_props["$ai_http_status"] == 200
+    assert first_call_props["$ai_input_tokens"] == 17
+    assert first_call_props["$ai_output_tokens"] is not None
+
+
+def test_tool_calls(mock_client):
+    prompt = ChatPromptTemplate.from_messages([("user", "Foo")])
+    model = FakeMessagesListChatModel(
+        responses=[
+            AIMessage(
+                content="Bar",
+                additional_kwargs={
+                    "tool_calls": [
+                        {
+                            "type": "function",
+                            "id": "123",
+                            "function": {
+                                "name": "test",
+                                "args": '{"a": 1}',
+                            },
+                        }
+                    ]
+                },
+            )
+        ]
+    )
+    chain = prompt | model
+    callbacks = CallbackHandler(mock_client)
+    chain.invoke({}, config={"callbacks": [callbacks]})
+
+    assert mock_client.capture.call_count == 1
+    call = mock_client.capture.call_args[1]
+    assert call["properties"]["$ai_output_choices"][0]["tool_calls"] == [
+        {
+            "type": "function",
+            "id": "123",
+            "function": {
+                "name": "test",
+                "args": '{"a": 1}',
+            },
+        }
+    ]
+    assert "additional_kwargs" not in call["properties"]["$ai_output_choices"][0]

posthog/test/ai/openai/test_openai.py

@@ -0,0 +1,175 @@
+import time
+from unittest.mock import patch
+
+import pytest
+from openai.types.chat import ChatCompletion, ChatCompletionMessage
+from openai.types.chat.chat_completion import Choice
+from openai.types.completion_usage import CompletionUsage
+from openai.types.create_embedding_response import CreateEmbeddingResponse, Usage
+from openai.types.embedding import Embedding
+
+from posthog.ai.openai import OpenAI
+
+
+@pytest.fixture
+def mock_client():
+    with patch("posthog.client.Client") as mock_client:
+        mock_client.privacy_mode = False
+        yield mock_client
+
+
+@pytest.fixture
+def mock_openai_response():
+    return ChatCompletion(
+        id="test",
+        model="gpt-4",
+        object="chat.completion",
+        created=int(time.time()),
+        choices=[
+            Choice(
+                finish_reason="stop",
+                index=0,
+                message=ChatCompletionMessage(
+                    content="Test response",
+                    role="assistant",
+                ),
+            )
+        ],
+        usage=CompletionUsage(
+            completion_tokens=10,
+            prompt_tokens=20,
+            total_tokens=30,
+        ),
+    )
+
+
+@pytest.fixture
+def mock_embedding_response():
+    return CreateEmbeddingResponse(
+        data=[
+            Embedding(
+                embedding=[0.1, 0.2, 0.3],
+                index=0,
+                object="embedding",
+            )
+        ],
+        model="text-embedding-3-small",
+        object="list",
+        usage=Usage(
+            prompt_tokens=10,
+            total_tokens=10,
+        ),
+    )
+
+
+def test_basic_completion(mock_client, mock_openai_response):
+    with patch("openai.resources.chat.completions.Completions.create", return_value=mock_openai_response):
+        client = OpenAI(api_key="test-key", posthog_client=mock_client)
+        response = client.chat.completions.create(
+            model="gpt-4",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_properties={"foo": "bar"},
+        )
+
+        assert response == mock_openai_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["$ai_model"] == "gpt-4"
+        assert props["$ai_input"] == [{"role": "user", "content": "Hello"}]
+        assert props["$ai_output_choices"] == [{"role": "assistant", "content": "Test response"}]
+        assert props["$ai_input_tokens"] == 20
+        assert props["$ai_output_tokens"] == 10
+        assert props["$ai_http_status"] == 200
+        assert props["foo"] == "bar"
+        assert isinstance(props["$ai_latency"], float)
+
+
+def test_embeddings(mock_client, mock_embedding_response):
+    with patch("openai.resources.embeddings.Embeddings.create", return_value=mock_embedding_response):
+        client = OpenAI(api_key="test-key", posthog_client=mock_client)
+        response = client.embeddings.create(
+            model="text-embedding-3-small",
+            input="Hello world",
+            posthog_distinct_id="test-id",
+            posthog_properties={"foo": "bar"},
+        )
+
+        assert response == mock_embedding_response
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+
+        assert call_args["distinct_id"] == "test-id"
+        assert call_args["event"] == "$ai_embedding"
+        assert props["$ai_provider"] == "openai"
+        assert props["$ai_model"] == "text-embedding-3-small"
+        assert props["$ai_input"] == "Hello world"
+        assert props["$ai_input_tokens"] == 10
+        assert props["$ai_http_status"] == 200
+        assert props["foo"] == "bar"
+        assert isinstance(props["$ai_latency"], float)
+
+
+def test_groups(mock_client, mock_openai_response):
+    with patch("openai.resources.chat.completions.Completions.create", return_value=mock_openai_response):
+        client = OpenAI(api_key="test-key", posthog_client=mock_client)
+        response = client.chat.completions.create(
+            model="gpt-4",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_groups={"company": "test_company"},
+        )
+
+        assert response == mock_openai_response
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+
+        assert call_args["groups"] == {"company": "test_company"}
+
+
+def test_privacy_mode_local(mock_client, mock_openai_response):
+    with patch("openai.resources.chat.completions.Completions.create", return_value=mock_openai_response):
+        client = OpenAI(api_key="test-key", posthog_client=mock_client)
+        response = client.chat.completions.create(
+            model="gpt-4",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_privacy_mode=True,
+        )
+
+        assert response == mock_openai_response
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+        assert props["$ai_input"] is None
+        assert props["$ai_output_choices"] is None
+
+
+def test_privacy_mode_global(mock_client, mock_openai_response):
+    with patch("openai.resources.chat.completions.Completions.create", return_value=mock_openai_response):
+        mock_client.privacy_mode = True
+        client = OpenAI(api_key="test-key", posthog_client=mock_client)
+        response = client.chat.completions.create(
+            model="gpt-4",
+            messages=[{"role": "user", "content": "Hello"}],
+            posthog_distinct_id="test-id",
+            posthog_privacy_mode=False,
+        )
+
+        assert response == mock_openai_response
+        assert mock_client.capture.call_count == 1
+
+        call_args = mock_client.capture.call_args[1]
+        props = call_args["properties"]
+        assert props["$ai_input"] is None
+        assert props["$ai_output_choices"] is None

posthog/test/exception_integrations/__init__.py

@@ -0,0 +1,3 @@
+import pytest
+
+pytest.importorskip("django")

posthog/test/exception_integrations/test_django.py

@@ -0,0 +1,68 @@
+from posthog.exception_integrations.django import DjangoRequestExtractor
+
+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"
+)
+
+
+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 = {
+            "User-Agent": DEFAULT_USER_AGENT,
+            "Referrer": "http://example.com",
+            "X-Forwarded-For": "193.4.5.12",
+            **(override_headers or {}),
+        }
+
+    return Request()
+
+
+def test_request_extractor_with_no_trace():
+    request = mock_request_factory(None)
+    extractor = DjangoRequestExtractor(request)
+    assert extractor.extract_person_data() == {
+        "ip": "193.4.5.12",
+        "user_agent": DEFAULT_USER_AGENT,
+        "traceparent": None,
+        "distinct_id": None,
+    }
+
+
+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,
+    }
+
+
+def test_request_extractor_with_tracestate():
+    request = mock_request_factory(
+        {
+            "traceparent": "00-0af7651916cd43dd8448eb211c80319c-b7ad6b7169203331-01",
+            "tracestate": "posthog-distinct-id=1234",
+        }
+    )
+    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": "1234",
+    }
+
+
+def test_request_extractor_with_complicated_tracestate():
+    request = mock_request_factory({"tracestate": "posthog-distinct-id=alohaMountainsXUYZ,rojo=00f067aa0ba902b7"})
+    extractor = DjangoRequestExtractor(request)
+    assert extractor.extract_person_data() == {
+        "ip": "193.4.5.12",
+        "user_agent": DEFAULT_USER_AGENT,
+        "traceparent": None,
+        "distinct_id": "alohaMountainsXUYZ",
+    }

posthog/test/test_client.py

@@ -84,6 +84,184 @@ class TestClient(unittest.TestCase):
         self.assertEqual(msg["properties"]["$lib"], "posthog-python")
         self.assertEqual(msg["properties"]["$lib_version"], VERSION)
 
+    def test_basic_super_properties(self):
+        client = Client(FAKE_TEST_API_KEY, super_properties={"source": "repo-name"})
+
+        _, msg = client.capture("distinct_id", "python test event")
+        client.flush()
+
+        self.assertEqual(msg["event"], "python test event")
+        self.assertEqual(msg["properties"]["source"], "repo-name")
+
+        _, msg = client.identify("distinct_id", {"trait": "value"})
+        client.flush()
+
+        self.assertEqual(msg["$set"]["trait"], "value")
+        self.assertEqual(msg["properties"]["source"], "repo-name")
+
+    def test_basic_capture_exception(self):
+
+        with mock.patch.object(Client, "capture", return_value=None) as patch_capture:
+            client = self.client
+            exception = Exception("test exception")
+            client.capture_exception(exception, distinct_id="distinct_id")
+
+            self.assertTrue(patch_capture.called)
+            capture_call = patch_capture.call_args[0]
+            self.assertEqual(capture_call[0], "distinct_id")
+            self.assertEqual(capture_call[1], "$exception")
+            self.assertEqual(
+                capture_call[2],
+                {
+                    "$exception_type": "Exception",
+                    "$exception_message": "test exception",
+                    "$exception_list": [
+                        {
+                            "mechanism": {"type": "generic", "handled": True},
+                            "module": None,
+                            "type": "Exception",
+                            "value": "test exception",
+                        }
+                    ],
+                    "$exception_personURL": "https://us.i.posthog.com/project/random_key/person/distinct_id",
+                },
+            )
+
+    def test_basic_capture_exception_with_distinct_id(self):
+
+        with mock.patch.object(Client, "capture", return_value=None) as patch_capture:
+            client = self.client
+            exception = Exception("test exception")
+            client.capture_exception(exception, "distinct_id")
+
+            self.assertTrue(patch_capture.called)
+            capture_call = patch_capture.call_args[0]
+            self.assertEqual(capture_call[0], "distinct_id")
+            self.assertEqual(capture_call[1], "$exception")
+            self.assertEqual(
+                capture_call[2],
+                {
+                    "$exception_type": "Exception",
+                    "$exception_message": "test exception",
+                    "$exception_list": [
+                        {
+                            "mechanism": {"type": "generic", "handled": True},
+                            "module": None,
+                            "type": "Exception",
+                            "value": "test exception",
+                        }
+                    ],
+                    "$exception_personURL": "https://us.i.posthog.com/project/random_key/person/distinct_id",
+                },
+            )
+
+    def test_basic_capture_exception_with_correct_host_generation(self):
+
+        with mock.patch.object(Client, "capture", return_value=None) as patch_capture:
+            client = Client(FAKE_TEST_API_KEY, on_error=self.set_fail, host="https://aloha.com")
+            exception = Exception("test exception")
+            client.capture_exception(exception, "distinct_id")
+
+            self.assertTrue(patch_capture.called)
+            capture_call = patch_capture.call_args[0]
+            self.assertEqual(capture_call[0], "distinct_id")
+            self.assertEqual(capture_call[1], "$exception")
+            self.assertEqual(
+                capture_call[2],
+                {
+                    "$exception_type": "Exception",
+                    "$exception_message": "test exception",
+                    "$exception_list": [
+                        {
+                            "mechanism": {"type": "generic", "handled": True},
+                            "module": None,
+                            "type": "Exception",
+                            "value": "test exception",
+                        }
+                    ],
+                    "$exception_personURL": "https://aloha.com/project/random_key/person/distinct_id",
+                },
+            )
+
+    def test_basic_capture_exception_with_correct_host_generation_for_server_hosts(self):
+
+        with mock.patch.object(Client, "capture", return_value=None) as patch_capture:
+            client = Client(FAKE_TEST_API_KEY, on_error=self.set_fail, host="https://app.posthog.com")
+            exception = Exception("test exception")
+            client.capture_exception(exception, "distinct_id")
+
+            self.assertTrue(patch_capture.called)
+            capture_call = patch_capture.call_args[0]
+            self.assertEqual(capture_call[0], "distinct_id")
+            self.assertEqual(capture_call[1], "$exception")
+            self.assertEqual(
+                capture_call[2],
+                {
+                    "$exception_type": "Exception",
+                    "$exception_message": "test exception",
+                    "$exception_list": [
+                        {
+                            "mechanism": {"type": "generic", "handled": True},
+                            "module": None,
+                            "type": "Exception",
+                            "value": "test exception",
+                        }
+                    ],
+                    "$exception_personURL": "https://app.posthog.com/project/random_key/person/distinct_id",
+                },
+            )
+
+    def test_basic_capture_exception_with_no_exception_given(self):
+
+        with mock.patch.object(Client, "capture", return_value=None) as patch_capture:
+            client = self.client
+            try:
+                raise Exception("test exception")
+            except Exception:
+                client.capture_exception(distinct_id="distinct_id")
+
+            self.assertTrue(patch_capture.called)
+            capture_call = patch_capture.call_args[0]
+            self.assertEqual(capture_call[0], "distinct_id")
+            self.assertEqual(capture_call[1], "$exception")
+            self.assertEqual(capture_call[2]["$exception_type"], "Exception")
+            self.assertEqual(capture_call[2]["$exception_message"], "test exception")
+            self.assertEqual(capture_call[2]["$exception_list"][0]["mechanism"]["type"], "generic")
+            self.assertEqual(capture_call[2]["$exception_list"][0]["mechanism"]["handled"], True)
+            self.assertEqual(capture_call[2]["$exception_list"][0]["module"], None)
+            self.assertEqual(capture_call[2]["$exception_list"][0]["type"], "Exception")
+            self.assertEqual(capture_call[2]["$exception_list"][0]["value"], "test exception")
+            self.assertEqual(
+                capture_call[2]["$exception_list"][0]["stacktrace"]["type"],
+                "raw",
+            )
+            self.assertEqual(
+                capture_call[2]["$exception_list"][0]["stacktrace"]["frames"][0]["filename"],
+                "posthog/test/test_client.py",
+            )
+            self.assertEqual(
+                capture_call[2]["$exception_list"][0]["stacktrace"]["frames"][0]["function"],
+                "test_basic_capture_exception_with_no_exception_given",
+            )
+            self.assertEqual(
+                capture_call[2]["$exception_list"][0]["stacktrace"]["frames"][0]["module"], "posthog.test.test_client"
+            )
+            self.assertEqual(capture_call[2]["$exception_list"][0]["stacktrace"]["frames"][0]["in_app"], True)
+
+    def test_basic_capture_exception_with_no_exception_happening(self):
+
+        with mock.patch.object(Client, "capture", return_value=None) as patch_capture:
+            with self.assertLogs("posthog", level="WARNING") as logs:
+
+                client = self.client
+                client.capture_exception()
+
+                self.assertFalse(patch_capture.called)
+                self.assertEqual(
+                    logs.output[0],
+                    "WARNING:posthog:No exception information available",
+                )
+
     @mock.patch("posthog.client.decide")
     def test_basic_capture_with_feature_flags(self, patch_decide):
         patch_decide.return_value = {"featureFlags": {"beta-feature": "random-variant"}}
@@ -314,8 +492,8 @@ class TestClient(unittest.TestCase):
         self.assertEqual(patch_decide.call_count, 1)
         patch_decide.assert_called_with(
             "random_key",
-            None,
-            timeout=10,
+            "https://us.i.posthog.com",
+            timeout=3,
             distinct_id="distinct_id",
             groups={},
             person_properties=None,
@@ -330,7 +508,12 @@ class TestClient(unittest.TestCase):
         }
 
         client = Client(
-            FAKE_TEST_API_KEY, on_error=self.set_fail, personal_api_key=FAKE_TEST_API_KEY, disable_geoip=True
+            FAKE_TEST_API_KEY,
+            host="https://app.posthog.com",
+            on_error=self.set_fail,
+            personal_api_key=FAKE_TEST_API_KEY,
+            disable_geoip=True,
+            feature_flags_request_timeout_seconds=12,
         )
         success, msg = client.capture("distinct_id", "python test event", send_feature_flags=True, disable_geoip=False)
         client.flush()
@@ -351,8 +534,8 @@ class TestClient(unittest.TestCase):
         self.assertEqual(patch_decide.call_count, 1)
         patch_decide.assert_called_with(
             "random_key",
-            None,
-            timeout=10,
+            "https://us.i.posthog.com",
+            timeout=12,
             distinct_id="distinct_id",
             groups={},
             person_properties=None,
@@ -532,6 +715,25 @@ class TestClient(unittest.TestCase):
         self.assertTrue(isinstance(msg["timestamp"], str))
         self.assertIsNone(msg.get("uuid"))
 
+    def test_basic_group_identify_with_distinct_id(self):
+        success, msg = self.client.group_identify("organization", "id:5", distinct_id="distinct_id")
+        self.assertTrue(success)
+        self.assertEqual(msg["event"], "$groupidentify")
+        self.assertEqual(msg["distinct_id"], "distinct_id")
+        self.assertEqual(
+            msg["properties"],
+            {
+                "$group_type": "organization",
+                "$group_key": "id:5",
+                "$group_set": {},
+                "$lib": "posthog-python",
+                "$lib_version": VERSION,
+                "$geoip_disable": True,
+            },
+        )
+        self.assertTrue(isinstance(msg["timestamp"], str))
+        self.assertIsNone(msg.get("uuid"))
+
     def test_advanced_group_identify(self):
         success, msg = self.client.group_identify(
             "organization", "id:5", {"trait": "value"}, {"ip": "192.168.0.1"}, datetime(2014, 9, 3), "new-uuid"
@@ -554,6 +756,35 @@ class TestClient(unittest.TestCase):
         self.assertEqual(msg["timestamp"], "2014-09-03T00:00:00+00:00")
         self.assertEqual(msg["context"]["ip"], "192.168.0.1")
 
+    def test_advanced_group_identify_with_distinct_id(self):
+        success, msg = self.client.group_identify(
+            "organization",
+            "id:5",
+            {"trait": "value"},
+            {"ip": "192.168.0.1"},
+            datetime(2014, 9, 3),
+            "new-uuid",
+            distinct_id="distinct_id",
+        )
+
+        self.assertTrue(success)
+        self.assertEqual(msg["event"], "$groupidentify")
+        self.assertEqual(msg["distinct_id"], "distinct_id")
+
+        self.assertEqual(
+            msg["properties"],
+            {
+                "$group_type": "organization",
+                "$group_key": "id:5",
+                "$group_set": {"trait": "value"},
+                "$lib": "posthog-python",
+                "$lib_version": VERSION,
+                "$geoip_disable": True,
+            },
+        )
+        self.assertEqual(msg["timestamp"], "2014-09-03T00:00:00+00:00")
+        self.assertEqual(msg["context"]["ip"], "192.168.0.1")
+
     def test_basic_alias(self):
         client = self.client
         success, msg = client.alias("previousId", "distinct_id")
@@ -779,11 +1010,11 @@ class TestClient(unittest.TestCase):
         client.get_feature_flag("random_key", "some_id", disable_geoip=True)
         patch_decide.assert_called_with(
             "random_key",
-            None,
-            timeout=10,
+            "https://us.i.posthog.com",
+            timeout=3,
             distinct_id="some_id",
             groups={},
-            person_properties={"$current_distinct_id": "some_id"},
+            person_properties={"distinct_id": "some_id"},
             group_properties={},
             disable_geoip=True,
         )
@@ -791,11 +1022,11 @@ class TestClient(unittest.TestCase):
         client.feature_enabled("random_key", "feature_enabled_distinct_id", disable_geoip=True)
         patch_decide.assert_called_with(
             "random_key",
-            None,
-            timeout=10,
+            "https://us.i.posthog.com",
+            timeout=3,
             distinct_id="feature_enabled_distinct_id",
             groups={},
-            person_properties={"$current_distinct_id": "feature_enabled_distinct_id"},
+            person_properties={"distinct_id": "feature_enabled_distinct_id"},
             group_properties={},
             disable_geoip=True,
         )
@@ -803,11 +1034,11 @@ class TestClient(unittest.TestCase):
         client.get_all_flags_and_payloads("all_flags_payloads_id")
         patch_decide.assert_called_with(
             "random_key",
-            None,
-            timeout=10,
+            "https://us.i.posthog.com",
+            timeout=3,
             distinct_id="all_flags_payloads_id",
             groups={},
-            person_properties={"$current_distinct_id": "all_flags_payloads_id"},
+            person_properties={"distinct_id": "all_flags_payloads_id"},
             group_properties={},
             disable_geoip=False,
         )
@@ -829,7 +1060,7 @@ class TestClient(unittest.TestCase):
         patch_decide.return_value = {
             "featureFlags": {"beta-feature": "random-variant", "alpha-feature": True, "off-feature": False}
         }
-        client = Client(FAKE_TEST_API_KEY, on_error=self.set_fail, disable_geoip=False)
+        client = Client(FAKE_TEST_API_KEY, host="http://app2.posthog.com", on_error=self.set_fail, disable_geoip=False)
         client.get_feature_flag(
             "random_key",
             "some_id",
@@ -839,11 +1070,11 @@ class TestClient(unittest.TestCase):
         )
         patch_decide.assert_called_with(
             "random_key",
-            None,
-            timeout=10,
+            "http://app2.posthog.com",
+            timeout=3,
             distinct_id="some_id",
             groups={"company": "id:5", "instance": "app.posthog.com"},
-            person_properties={"$current_distinct_id": "some_id", "x1": "y1"},
+            person_properties={"distinct_id": "some_id", "x1": "y1"},
             group_properties={
                 "company": {"$group_key": "id:5", "x": "y"},
                 "instance": {"$group_key": "app.posthog.com"},
@@ -856,7 +1087,7 @@ class TestClient(unittest.TestCase):
             "random_key",
             "some_id",
             groups={"company": "id:5", "instance": "app.posthog.com"},
-            person_properties={"$current_distinct_id": "override"},
+            person_properties={"distinct_id": "override"},
             group_properties={
                 "company": {
                     "$group_key": "group_override",
@@ -865,11 +1096,11 @@ class TestClient(unittest.TestCase):
         )
         patch_decide.assert_called_with(
             "random_key",
-            None,
-            timeout=10,
+            "http://app2.posthog.com",
+            timeout=3,
             distinct_id="some_id",
             groups={"company": "id:5", "instance": "app.posthog.com"},
-            person_properties={"$current_distinct_id": "override"},
+            person_properties={"distinct_id": "override"},
             group_properties={
                 "company": {"$group_key": "group_override"},
                 "instance": {"$group_key": "app.posthog.com"},
@@ -882,11 +1113,11 @@ class TestClient(unittest.TestCase):
         client.get_all_flags_and_payloads("some_id", groups={}, person_properties=None, group_properties=None)
         patch_decide.assert_called_with(
             "random_key",
-            None,
-            timeout=10,
+            "http://app2.posthog.com",
+            timeout=3,
             distinct_id="some_id",
             groups={},
-            person_properties={"$current_distinct_id": "some_id"},
+            person_properties={"distinct_id": "some_id"},
             group_properties={},
             disable_geoip=False,
         )

posthog/test/test_exception_capture.py

@@ -0,0 +1,63 @@
+import subprocess
+import sys
+from textwrap import dedent
+
+import pytest
+
+
+def test_excepthook(tmpdir):
+    app = tmpdir.join("app.py")
+    app.write(
+        dedent(
+            """
+    from posthog import Posthog
+    posthog = Posthog('phc_x', host='https://eu.i.posthog.com', enable_exception_autocapture=True, debug=True, on_error=lambda e, batch: print('error handling batch: ', e, batch))
+
+    # frame_value = "LOL"
+
+    1/0
+    """
+        )
+    )
+
+    with pytest.raises(subprocess.CalledProcessError) as excinfo:
+        subprocess.check_output([sys.executable, str(app)], stderr=subprocess.STDOUT)
+
+    output = excinfo.value.output
+
+    assert b"ZeroDivisionError" in output
+    assert b"LOL" in output
+    assert b"DEBUG:posthog:data uploaded successfully" in output
+    assert (
+        b'"$exception_list": [{"mechanism": {"type": "generic", "handled": true}, "module": null, "type": "ZeroDivisionError", "value": "division by zero", "stacktrace": {"frames": [{"platform": "python", "filename": "app.py", "abs_path"'
+        in output
+    )
+
+
+def test_trying_to_use_django_integration(tmpdir):
+    app = tmpdir.join("app.py")
+    app.write(
+        dedent(
+            """
+    from posthog import Posthog, Integrations
+    posthog = Posthog('phc_x', host='https://eu.i.posthog.com', enable_exception_autocapture=True, exception_autocapture_integrations=[Integrations.Django], debug=True, on_error=lambda e, batch: print('error handling batch: ', e, batch))
+
+    # frame_value = "LOL"
+
+    1/0
+    """
+        )
+    )
+
+    with pytest.raises(subprocess.CalledProcessError) as excinfo:
+        subprocess.check_output([sys.executable, str(app)], stderr=subprocess.STDOUT)
+
+    output = excinfo.value.output
+
+    assert b"ZeroDivisionError" in output
+    assert b"LOL" in output
+    assert b"DEBUG:posthog:data uploaded successfully" in output
+    assert (
+        b'"$exception_list": [{"mechanism": {"type": "generic", "handled": true}, "module": null, "type": "ZeroDivisionError", "value": "division by zero", "stacktrace": {"frames": [{"platform": "python", "filename": "app.py", "abs_path"'
+        in output
+    )

posthog/test/test_feature_flags.py

@@ -960,6 +960,60 @@ class TestLocalEvaluation(unittest.TestCase):
         self.assertEqual(patch_decide.call_count, 0)
         self.assertEqual(patch_capture.call_count, 0)
 
+    @mock.patch("posthog.client.decide")
+    @mock.patch("posthog.client.get")
+    def test_feature_flags_local_evaluation_None_values(self, patch_get, patch_decide):
+        client = Client(FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+        client.feature_flags = [
+            {
+                id: 1,
+                "name": "Beta Feature",
+                "key": "beta-feature",
+                "is_simple_flag": True,
+                "active": True,
+                "filters": {
+                    "groups": [
+                        {
+                            "variant": None,
+                            "properties": [
+                                {"key": "latestBuildVersion", "type": "person", "value": ".+", "operator": "regex"},
+                                {"key": "latestBuildVersionMajor", "type": "person", "value": "23", "operator": "gt"},
+                                {"key": "latestBuildVersionMinor", "type": "person", "value": "31", "operator": "gt"},
+                                {"key": "latestBuildVersionPatch", "type": "person", "value": "0", "operator": "gt"},
+                            ],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            },
+        ]
+
+        feature_flag_match = client.get_feature_flag(
+            "beta-feature",
+            "some-distinct-id",
+            person_properties={
+                "latestBuildVersion": None,
+                "latestBuildVersionMajor": None,
+                "latestBuildVersionMinor": None,
+                "latestBuildVersionPatch": None,
+            },
+        )
+
+        self.assertEqual(feature_flag_match, False)
+        self.assertEqual(patch_decide.call_count, 0)
+        self.assertEqual(patch_get.call_count, 0)
+
+        feature_flag_match = client.get_feature_flag(
+            "beta-feature",
+            "some-distinct-id",
+            person_properties={
+                "latestBuildVersion": "24.32..1",
+                "latestBuildVersionMajor": "24",
+                "latestBuildVersionMinor": "32",
+                "latestBuildVersionPatch": "1",
+            },
+        )
+
     @mock.patch("posthog.client.decide")
     @mock.patch("posthog.client.get")
     def test_feature_flags_local_evaluation_for_cohorts(self, patch_get, patch_decide):
@@ -1578,9 +1632,10 @@ class TestLocalEvaluation(unittest.TestCase):
         )
         self.assertEqual(patch_decide.call_count, 0)
 
+    @mock.patch.object(Client, "capture")
     @mock.patch("posthog.client.decide")
-    def test_boolean_feature_flag_payload_decide(self, patch_decide):
-        patch_decide.return_value = {"featureFlagPayloads": {"person-flag": 300}}
+    def test_boolean_feature_flag_payload_decide(self, patch_decide, patch_capture):
+        patch_decide.return_value = {"featureFlags": {"person-flag": True}, "featureFlagPayloads": {"person-flag": 300}}
         self.assertEqual(
             self.client.get_feature_flag_payload(
                 "person-flag", "some-distinct-id", person_properties={"region": "USA"}
@@ -1595,6 +1650,8 @@ class TestLocalEvaluation(unittest.TestCase):
             300,
         )
         self.assertEqual(patch_decide.call_count, 2)
+        self.assertEqual(patch_capture.call_count, 1)
+        patch_capture.reset_mock()
 
     @mock.patch("posthog.client.decide")
     def test_multivariate_feature_flag_payloads(self, patch_decide):
@@ -1714,7 +1771,7 @@ class TestMatchProperties(unittest.TestCase):
         self.assertTrue(match_property(property_a, {"key": "value"}))
         self.assertTrue(match_property(property_a, {"key": "value2"}))
         self.assertTrue(match_property(property_a, {"key": ""}))
-        self.assertTrue(match_property(property_a, {"key": None}))
+        self.assertFalse(match_property(property_a, {"key": None}))
 
         with self.assertRaises(InconclusiveMatchError):
             match_property(property_a, {"key2": "value"})
@@ -1873,7 +1930,7 @@ class TestMatchProperties(unittest.TestCase):
 
     @freeze_time("2022-05-01")
     def test_match_property_relative_date_operators(self):
-        property_a = self.property(key="key", value="6h", operator="is_relative_date_before")
+        property_a = self.property(key="key", value="-6h", operator="is_date_before")
         self.assertTrue(match_property(property_a, {"key": "2022-03-01"}))
         self.assertTrue(match_property(property_a, {"key": "2022-04-30"}))
         self.assertTrue(match_property(property_a, {"key": datetime.datetime(2022, 4, 30, 1, 2, 3)}))
@@ -1898,7 +1955,7 @@ class TestMatchProperties(unittest.TestCase):
         with self.assertRaises(InconclusiveMatchError):
             match_property(property_a, {"key": "abcdef"})
 
-        property_b = self.property(key="key", value="1h", operator="is_relative_date_after")
+        property_b = self.property(key="key", value="1h", operator="is_date_after")
         self.assertTrue(match_property(property_b, {"key": "2022-05-02"}))
         self.assertTrue(match_property(property_b, {"key": "2022-05-30"}))
         self.assertTrue(match_property(property_b, {"key": datetime.datetime(2022, 5, 30)}))
@@ -1910,16 +1967,16 @@ class TestMatchProperties(unittest.TestCase):
             self.assertFalse(match_property(property_b, {"key": "abcdef"}))
 
         # Invalid flag property
-        property_c = self.property(key="key", value=1234, operator="is_relative_date_after")
+        property_c = self.property(key="key", value=1234, operator="is_date_after")
 
         with self.assertRaises(InconclusiveMatchError):
             self.assertFalse(match_property(property_c, {"key": 1}))
 
-        with self.assertRaises(InconclusiveMatchError):
-            self.assertFalse(match_property(property_c, {"key": "2022-05-30"}))
+        # parsed as 1234-05-01 for some reason?
+        self.assertTrue(match_property(property_c, {"key": "2022-05-30"}))
 
         # # Timezone aware property
-        property_d = self.property(key="key", value="12d", operator="is_relative_date_before")
+        property_d = self.property(key="key", value="12d", operator="is_date_before")
         self.assertFalse(match_property(property_d, {"key": "2022-05-30"}))
 
         self.assertTrue(match_property(property_d, {"key": "2022-03-30"}))
@@ -1929,45 +1986,45 @@ class TestMatchProperties(unittest.TestCase):
         self.assertFalse(match_property(property_d, {"key": "2022-04-19 02:00:01+02:00"}))
 
         # Try all possible relative dates
-        property_e = self.property(key="key", value="1h", operator="is_relative_date_before")
+        property_e = self.property(key="key", value="1h", operator="is_date_before")
         self.assertFalse(match_property(property_e, {"key": "2022-05-01 00:00:00"}))
         self.assertTrue(match_property(property_e, {"key": "2022-04-30 22:00:00"}))
 
-        property_f = self.property(key="key", value="1d", operator="is_relative_date_before")
+        property_f = self.property(key="key", value="-1d", operator="is_date_before")
         self.assertTrue(match_property(property_f, {"key": "2022-04-29 23:59:00"}))
         self.assertFalse(match_property(property_f, {"key": "2022-04-30 00:00:01"}))
 
-        property_g = self.property(key="key", value="1w", operator="is_relative_date_before")
+        property_g = self.property(key="key", value="1w", operator="is_date_before")
         self.assertTrue(match_property(property_g, {"key": "2022-04-23 00:00:00"}))
         self.assertFalse(match_property(property_g, {"key": "2022-04-24 00:00:00"}))
         self.assertFalse(match_property(property_g, {"key": "2022-04-24 00:00:01"}))
 
-        property_h = self.property(key="key", value="1m", operator="is_relative_date_before")
+        property_h = self.property(key="key", value="1m", operator="is_date_before")
         self.assertTrue(match_property(property_h, {"key": "2022-03-01 00:00:00"}))
         self.assertFalse(match_property(property_h, {"key": "2022-04-05 00:00:00"}))
 
-        property_i = self.property(key="key", value="1y", operator="is_relative_date_before")
+        property_i = self.property(key="key", value="1y", operator="is_date_before")
         self.assertTrue(match_property(property_i, {"key": "2021-04-28 00:00:00"}))
         self.assertFalse(match_property(property_i, {"key": "2021-05-01 00:00:01"}))
 
-        property_j = self.property(key="key", value="122h", operator="is_relative_date_after")
+        property_j = self.property(key="key", value="122h", operator="is_date_after")
         self.assertTrue(match_property(property_j, {"key": "2022-05-01 00:00:00"}))
         self.assertFalse(match_property(property_j, {"key": "2022-04-23 01:00:00"}))
 
-        property_k = self.property(key="key", value="2d", operator="is_relative_date_after")
+        property_k = self.property(key="key", value="2d", operator="is_date_after")
         self.assertTrue(match_property(property_k, {"key": "2022-05-01 00:00:00"}))
         self.assertTrue(match_property(property_k, {"key": "2022-04-29 00:00:01"}))
         self.assertFalse(match_property(property_k, {"key": "2022-04-29 00:00:00"}))
 
-        property_l = self.property(key="key", value="02w", operator="is_relative_date_after")
+        property_l = self.property(key="key", value="-02w", operator="is_date_after")
         self.assertTrue(match_property(property_l, {"key": "2022-05-01 00:00:00"}))
         self.assertFalse(match_property(property_l, {"key": "2022-04-16 00:00:00"}))
 
-        property_m = self.property(key="key", value="1m", operator="is_relative_date_after")
+        property_m = self.property(key="key", value="1m", operator="is_date_after")
         self.assertTrue(match_property(property_m, {"key": "2022-04-01 00:00:01"}))
         self.assertFalse(match_property(property_m, {"key": "2022-04-01 00:00:00"}))
 
-        property_n = self.property(key="key", value="1y", operator="is_relative_date_after")
+        property_n = self.property(key="key", value="1y", operator="is_date_after")
         self.assertTrue(match_property(property_n, {"key": "2022-05-01 00:00:00"}))
         self.assertTrue(match_property(property_n, {"key": "2021-05-01 00:00:01"}))
         self.assertFalse(match_property(property_n, {"key": "2021-05-01 00:00:00"}))
@@ -1980,20 +2037,20 @@ class TestMatchProperties(unittest.TestCase):
         self.assertTrue(match_property(property_a, {"key": "non"}))
 
         property_b = self.property(key="key", value=None, operator="is_set")
-        self.assertTrue(match_property(property_b, {"key": None}))
+        self.assertFalse(match_property(property_b, {"key": None}))
 
         property_c = self.property(key="key", value="no", operator="icontains")
-        self.assertTrue(match_property(property_c, {"key": None}))
+        self.assertFalse(match_property(property_c, {"key": None}))
         self.assertFalse(match_property(property_c, {"key": "smh"}))
 
         property_d = self.property(key="key", value="No", operator="regex")
-        self.assertTrue(match_property(property_d, {"key": None}))
+        self.assertFalse(match_property(property_d, {"key": None}))
 
         property_d_lower_case = self.property(key="key", value="no", operator="regex")
         self.assertFalse(match_property(property_d_lower_case, {"key": None}))
 
         property_e = self.property(key="key", value=1, operator="gt")
-        self.assertTrue(match_property(property_e, {"key": None}))
+        self.assertFalse(match_property(property_e, {"key": None}))
 
         property_f = self.property(key="key", value=1, operator="lt")
         self.assertFalse(match_property(property_f, {"key": None}))
@@ -2002,15 +2059,13 @@ class TestMatchProperties(unittest.TestCase):
         self.assertFalse(match_property(property_g, {"key": None}))
 
         property_h = self.property(key="key", value="Oo", operator="lte")
-        self.assertTrue(match_property(property_h, {"key": None}))
+        self.assertFalse(match_property(property_h, {"key": None}))
 
         property_i = self.property(key="key", value="2022-05-01", operator="is_date_before")
-        with self.assertRaises(InconclusiveMatchError):
-            self.assertFalse(match_property(property_i, {"key": None}))
+        self.assertFalse(match_property(property_i, {"key": None}))
 
         property_j = self.property(key="key", value="2022-05-01", operator="is_date_after")
-        with self.assertRaises(InconclusiveMatchError):
-            self.assertFalse(match_property(property_j, {"key": None}))
+        self.assertFalse(match_property(property_j, {"key": None}))
 
         property_k = self.property(key="key", value="2022-05-01", operator="is_date_before")
         with self.assertRaises(InconclusiveMatchError):
@@ -2282,6 +2337,88 @@ class TestCaptureCalls(unittest.TestCase):
             disable_geoip=None,
         )
 
+    @mock.patch.object(Client, "capture")
+    @mock.patch("posthog.client.decide")
+    def test_capture_is_called_in_get_feature_flag_payload(self, patch_decide, patch_capture):
+        patch_decide.return_value = {
+            "featureFlags": {"person-flag": True},
+            "featureFlagPayloads": {"person-flag": 300},
+        }
+        client = Client(api_key=FAKE_TEST_API_KEY, personal_api_key=FAKE_TEST_API_KEY)
+
+        client.feature_flags = [
+            {
+                "id": 1,
+                "name": "Beta Feature",
+                "key": "person-flag",
+                "is_simple_flag": False,
+                "active": True,
+                "filters": {
+                    "groups": [
+                        {
+                            "properties": [{"key": "region", "value": "USA"}],
+                            "rollout_percentage": 100,
+                        }
+                    ],
+                },
+            }
+        ]
+
+        # Call get_feature_flag_payload with match_value=None to trigger get_feature_flag
+        client.get_feature_flag_payload(
+            key="person-flag", distinct_id="some-distinct-id", person_properties={"region": "USA", "name": "Aloha"}
+        )
+
+        # Assert that capture was called once, with the correct parameters
+        self.assertEqual(patch_capture.call_count, 1)
+        patch_capture.assert_called_with(
+            "some-distinct-id",
+            "$feature_flag_called",
+            {
+                "$feature_flag": "person-flag",
+                "$feature_flag_response": True,
+                "$feature_flag_payload": 300,
+                "locally_evaluated": False,
+                "$feature/person-flag": True,
+            },
+            groups={},
+            disable_geoip=None,
+        )
+
+        # Reset mocks for further tests
+        patch_capture.reset_mock()
+        patch_decide.reset_mock()
+
+        # Call get_feature_flag_payload again for the same user; capture should not be called again because we've already reported an event for this distinct_id + flag
+        client.get_feature_flag_payload(
+            key="person-flag", distinct_id="some-distinct-id", person_properties={"region": "USA", "name": "Aloha"}
+        )
+
+        self.assertEqual(patch_capture.call_count, 0)
+        patch_capture.reset_mock()
+
+        # Call get_feature_flag_payload for a different user; capture should be called
+        client.get_feature_flag_payload(
+            key="person-flag", distinct_id="some-distinct-id2", person_properties={"region": "USA", "name": "Aloha"}
+        )
+
+        self.assertEqual(patch_capture.call_count, 1)
+        patch_capture.assert_called_with(
+            "some-distinct-id2",
+            "$feature_flag_called",
+            {
+                "$feature_flag": "person-flag",
+                "$feature_flag_response": True,
+                "$feature_flag_payload": 300,
+                "locally_evaluated": False,
+                "$feature/person-flag": True,
+            },
+            groups={},
+            disable_geoip=None,
+        )
+
+        patch_capture.reset_mock()
+
     @mock.patch.object(Client, "capture")
     @mock.patch("posthog.client.decide")
     def test_disable_geoip_get_flag_capture_call(self, patch_decide, patch_capture):

posthog/test/test_request.py

@@ -2,9 +2,10 @@ import json
 import unittest
 from datetime import date, datetime
 
+import pytest
 import requests
 
-from posthog.request import DatetimeSerializer, batch_post
+from posthog.request import DatetimeSerializer, batch_post, determine_server_host
 from posthog.test.test_utils import TEST_API_KEY
 
 
@@ -42,3 +43,26 @@ class TestRequests(unittest.TestCase):
             batch_post(
                 "key", batch=[{"distinct_id": "distinct_id", "event": "python event", "type": "track"}], timeout=0.0001
             )
+
+
+@pytest.mark.parametrize(
+    "host, expected",
+    [
+        ("https://t.posthog.com", "https://t.posthog.com"),
+        ("https://t.posthog.com/", "https://t.posthog.com/"),
+        ("t.posthog.com", "t.posthog.com"),
+        ("t.posthog.com/", "t.posthog.com/"),
+        ("https://us.posthog.com.rg.proxy.com", "https://us.posthog.com.rg.proxy.com"),
+        ("app.posthog.com", "app.posthog.com"),
+        ("eu.posthog.com", "eu.posthog.com"),
+        ("https://app.posthog.com", "https://us.i.posthog.com"),
+        ("https://eu.posthog.com", "https://eu.i.posthog.com"),
+        ("https://us.posthog.com", "https://us.i.posthog.com"),
+        ("https://app.posthog.com/", "https://us.i.posthog.com"),
+        ("https://eu.posthog.com/", "https://eu.i.posthog.com"),
+        ("https://us.posthog.com/", "https://us.i.posthog.com"),
+        (None, "https://us.i.posthog.com"),
+    ],
+)
+def test_routing_to_custom_host(host, expected):
+    assert determine_server_host(host) == expected

posthog/version.py

@@ -1,4 +1,4 @@
-VERSION = "3.3.2"
+VERSION = "3.8.4"
 
 if __name__ == "__main__":
     print(VERSION, end="")  # noqa: T201

sentry_django_example/sentry_django_example/urls.py

@@ -13,6 +13,7 @@ Including another URLconf
     1. Import the include() function: from django.urls import include, path
     2. Add a URL to urlpatterns:  path('blog/', include('blog.urls'))
 """
+
 from django.contrib import admin
 from django.urls import path
 

setup.cfg

@@ -1,2 +1,5 @@
 [bdist_wheel]
 universal = 1
+
+[tool:pytest]
+asyncio_mode = auto

setup.py

@@ -14,7 +14,13 @@ long_description = """
 PostHog is developer-friendly, self-hosted product analytics. posthog-python is the python package.
 """
 
-install_requires = ["requests>=2.7,<3.0", "six>=1.5", "monotonic>=1.5", "backoff>=1.10.0", "python-dateutil>2.1"]
+install_requires = [
+    "requests>=2.7,<3.0",
+    "six>=1.5",
+    "monotonic>=1.5",
+    "backoff>=1.10.0",
+    "python-dateutil>2.1",
+]
 
 extras_require = {
     "dev": [
@@ -24,8 +30,24 @@ extras_require = {
         "flake8-print",
         "pre-commit",
     ],
-    "test": ["mock>=2.0.0", "freezegun==0.3.15", "pylint", "flake8", "coverage", "pytest", "pytest-timeout"],
+    "test": [
+        "mock>=2.0.0",
+        "freezegun==0.3.15",
+        "pylint",
+        "flake8",
+        "coverage",
+        "pytest",
+        "pytest-timeout",
+        "pytest-asyncio",
+        "django",
+        "openai",
+        "anthropic",
+        "langchain-community>=0.2.0",
+        "langchain-openai>=0.2.0",
+        "langchain-anthropic>=0.2.0",
+    ],
     "sentry": ["sentry-sdk", "django"],
+    "langchain": ["langchain>=0.2.0"],
 }
 
 setup(
@@ -37,7 +59,16 @@ setup(
     maintainer="PostHog",
     maintainer_email="hey@posthog.com",
     test_suite="posthog.test.all",
-    packages=["posthog", "posthog.test", "posthog.sentry"],
+    packages=[
+        "posthog",
+        "posthog.ai",
+        "posthog.ai.langchain",
+        "posthog.ai.openai",
+        "posthog.ai.anthropic",
+        "posthog.test",
+        "posthog.sentry",
+        "posthog.exception_integrations",
+    ],
     license="MIT License",
     install_requires=install_requires,
     extras_require=extras_require,
@@ -60,5 +91,8 @@ setup(
         "Programming Language :: Python :: 3.6",
         "Programming Language :: Python :: 3.7",
         "Programming Language :: Python :: 3.8",
+        "Programming Language :: Python :: 3.9",
+        "Programming Language :: Python :: 3.10",
+        "Programming Language :: Python :: 3.11",
     ],
 )

setup_analytics.py

@@ -27,7 +27,16 @@ setup(
     maintainer="PostHog",
     maintainer_email="hey@posthog.com",
     test_suite="posthoganalytics.test.all",
-    packages=["posthoganalytics", "posthoganalytics.test", "posthoganalytics.sentry"],
+    packages=[
+        "posthoganalytics",
+        "posthoganalytics.ai",
+        "posthoganalytics.ai.langchain",
+        "posthoganalytics.ai.openai",
+        "posthoganalytics.ai.anthropic",
+        "posthoganalytics.test",
+        "posthoganalytics.sentry",
+        "posthoganalytics.exception_integrations",
+    ],
     license="MIT License",
     install_requires=install_requires,
     tests_require=tests_require,
@@ -53,5 +62,8 @@ setup(
         "Programming Language :: Python :: 3.6",
         "Programming Language :: Python :: 3.7",
         "Programming Language :: Python :: 3.8",
+        "Programming Language :: Python :: 3.9",
+        "Programming Language :: Python :: 3.10",
+        "Programming Language :: Python :: 3.11",
     ],
 )