fix(ai): support async with on async streaming responses (Fixes #393)

.sampo/changesets/async-stream-context-manager.md

@@ -0,0 +1,5 @@
+---
+pypi/posthog: patch
+---
+
+Fix async streaming responses from the AI wrappers (OpenAI, Anthropic, Gemini) so they support `async with` as well as `async for`. Previously, consuming a stream via `async with` (e.g. with pydantic-ai) raised `TypeError: 'async_generator' object does not support the asynchronous context manager protocol`.

posthog/ai/anthropic/anthropic_async.py

@@ -11,6 +11,7 @@
 from typing import Any, Dict, List, Optional
 
 from posthog import setup
+from posthog.ai.stream import AsyncStreamWrapper
 from posthog.ai.types import StreamingContentBlock, TokenUsage, ToolInProgress
 from posthog.ai.utils import (
     call_llm_and_track_usage_async,
@@ -225,7 +226,7 @@ async def generator():
                     stop_reason=stop_reason,
                 )
 
-        return generator()
+        return AsyncStreamWrapper(generator(), stream=response)
 
     async def _capture_streaming_event(
         self,

posthog/ai/gemini/gemini_async.py

@@ -3,6 +3,7 @@
 import uuid
 from typing import Any, Dict, Optional
 
+from posthog.ai.stream import AsyncStreamWrapper
 from posthog.ai.types import TokenUsage, StreamingEventData
 from posthog.ai.utils import merge_system_prompt
 
@@ -354,7 +355,7 @@ async def async_generator():
                     stop_reason=stop_reason,
                 )
 
-        return async_generator()
+        return AsyncStreamWrapper(async_generator(), stream=response)
 
     def _capture_streaming_event(
         self,

posthog/ai/openai/openai_async.py

@@ -2,6 +2,7 @@
 import uuid
 from typing import Any, Dict, List, Optional
 
+from posthog.ai.stream import AsyncStreamWrapper
 from posthog.ai.types import TokenUsage
 
 try:
@@ -221,7 +222,7 @@ async def async_generator():
                     stop_reason=stop_reason,
                 )
 
-        return async_generator()
+        return AsyncStreamWrapper(async_generator(), stream=response)
 
     async def _capture_streaming_event(
         self,
@@ -515,7 +516,7 @@ async def async_generator():
                     stop_reason=stop_reason,
                 )
 
-        return async_generator()
+        return AsyncStreamWrapper(async_generator(), stream=response)
 
     async def _capture_streaming_event(
         self,

posthog/ai/stream.py

@@ -0,0 +1,62 @@
+"""Shared async streaming utilities for PostHog AI wrappers."""
+
+from typing import Any, AsyncGenerator, Generic, Optional, TypeVar
+
+T = TypeVar("T")
+
+
+class AsyncStreamWrapper(Generic[T]):
+    """Adds the async context manager protocol to a PostHog streaming generator.
+
+    The OpenAI and Anthropic SDK streams support both ``async for`` and
+    ``async with``. PostHog's wrappers returned a bare async generator, which
+    only supports ``async for``, so ``async with response:`` (used by
+    pydantic-ai) raised a TypeError. This wraps the tracking generator and,
+    when given the original provider stream, closes it and proxies attribute
+    access (e.g. ``.response``) to it.
+    """
+
+    def __init__(
+        self,
+        generator: AsyncGenerator[T, None],
+        stream: Optional[Any] = None,
+    ) -> None:
+        self._generator = generator
+        self._stream = stream
+
+    def __aiter__(self) -> "AsyncStreamWrapper[T]":
+        return self
+
+    async def __anext__(self) -> T:
+        return await self._generator.__anext__()
+
+    async def __aenter__(self) -> "AsyncStreamWrapper[T]":
+        return self
+
+    async def __aexit__(self, exc_type: Any, exc_val: Any, exc_tb: Any) -> bool:
+        # Close the generator first so its `finally` captures the event, even on
+        # early exit. try/finally still closes the provider stream if that raises.
+        try:
+            await self._generator.aclose()
+        finally:
+            if self._stream is not None:
+                close = getattr(self._stream, "aclose", None) or getattr(
+                    self._stream, "close", None
+                )
+                if close is not None:
+                    await close()
+
+        return False
+
+    # aclose/asend/athrow belong to the generator; provider streams expose
+    # close(), not these. Forwarding aclose() keeps it firing the event.
+    _GENERATOR_METHODS = ("aclose", "asend", "athrow")
+
+    def __getattr__(self, name: str) -> Any:
+        # Proxy only public attributes (e.g. `.response`) to the provider stream.
+        if name.startswith("_"):
+            raise AttributeError(name)
+        if name in self._GENERATOR_METHODS:
+            return getattr(self._generator, name)
+        target = self._stream if self._stream is not None else self._generator
+        return getattr(target, name)

posthog/test/ai/anthropic/test_anthropic.py

@@ -10,6 +10,7 @@
     from anthropic.types import Message, Usage
 
     from posthog.ai.anthropic import Anthropic, AsyncAnthropic
+    from posthog.test.ai.utils import RecordingAsyncStream
 
     ANTHROPIC_AVAILABLE = True
 except ImportError:
@@ -1421,3 +1422,74 @@ def test_integration_stop_reason(mock_client):
     assert props["$ai_stop_reason"] in ("end_turn", "max_tokens")
     assert props["$ai_provider"] == "anthropic"
     assert props["$ai_input_tokens"] > 0
+
+
+def _anthropic_stream_events():
+    final = MockStreamEvent("message_delta")
+    final.usage = MockUsage(
+        input_tokens=10,
+        output_tokens=5,
+        cache_read_input_tokens=0,
+        cache_creation_input_tokens=0,
+    )
+    return [
+        MockStreamEvent("message_start"),
+        MockStreamEvent("content_block_delta", text="Hi"),
+        final,
+    ]
+
+
+@pytest.mark.asyncio
+async def test_async_messages_create_streaming_supports_async_with(mock_client):
+    """Regression test for #393: messages.create(stream=True) must support
+    `async with`."""
+
+    async def mock_async_create(**kwargs):
+        return RecordingAsyncStream(_anthropic_stream_events())
+
+    with patch(
+        "anthropic.resources.messages.AsyncMessages.create",
+        side_effect=mock_async_create,
+    ):
+        client = AsyncAnthropic(posthog_client=mock_client)
+        response = await client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Foo"}],
+            stream=True,
+            max_tokens=1,
+        )
+
+        async with response as stream:
+            events = [event async for event in stream]
+
+    assert len(events) == 3
+    assert mock_client.capture.call_count == 1
+
+
+@pytest.mark.asyncio
+async def test_async_messages_streaming_early_exit_closes_provider_stream(mock_client):
+    """Breaking out early must close the underlying Anthropic stream and still
+    capture the event."""
+    source = RecordingAsyncStream(_anthropic_stream_events())
+
+    async def mock_async_create(**kwargs):
+        return source
+
+    with patch(
+        "anthropic.resources.messages.AsyncMessages.create",
+        side_effect=mock_async_create,
+    ):
+        client = AsyncAnthropic(posthog_client=mock_client)
+        response = await client.messages.create(
+            model="claude-3-opus-20240229",
+            messages=[{"role": "user", "content": "Foo"}],
+            stream=True,
+            max_tokens=1,
+        )
+
+        async with response as stream:
+            async for _ in stream:
+                break
+
+    assert source.closed is True
+    assert mock_client.capture.call_count == 1

posthog/test/ai/gemini/test_gemini_async.py

@@ -1110,3 +1110,43 @@ async def test_async_embed_content_integration_batch(mock_client):
 
     assert response.embeddings is not None
     assert len(response.embeddings) == len(inputs)
+
+
+async def test_async_client_streaming_supports_async_with(
+    mock_client, mock_google_genai_client
+):
+    """Regression test for #393: generate_content_stream must support `async with`."""
+
+    async def mock_streaming_response():
+        chunk = MagicMock()
+        chunk.text = "Hi"
+        usage = MagicMock()
+        usage.prompt_token_count = 5
+        usage.candidates_token_count = 3
+        usage.cached_content_token_count = 0
+        usage.thoughts_token_count = 0
+        chunk.usage_metadata = usage
+        yield chunk
+
+    mock_google_genai_client.aio.models.generate_content_stream = AsyncMock(
+        return_value=mock_streaming_response()
+    )
+
+    client = AsyncClient(api_key="test-key", posthog_client=mock_client)
+
+    response = await client.models.generate_content_stream(
+        model="gemini-2.0-flash",
+        contents=["Hi"],
+        posthog_distinct_id="test-id",
+    )
+
+    chunks = []
+    async with response as stream:
+        async for chunk in stream:
+            chunks.append(chunk)
+
+    assert len(chunks) == 1
+    assert mock_client.capture.call_count == 1
+    call_args = mock_client.capture.call_args[1]
+    assert call_args["event"] == "$ai_generation"
+    assert call_args["properties"]["$ai_provider"] == "gemini"

posthog/test/ai/openai/test_openai.py

@@ -39,6 +39,7 @@
     from posthog.ai.openai import OpenAI
     from posthog.ai.openai.openai_async import AsyncOpenAI
     from posthog.ai.openai.wrapper_utils import reset_fallback_warnings
+    from posthog.test.ai.utils import RecordingAsyncStream
 
     OPENAI_AVAILABLE = True
 except ImportError:
@@ -2352,3 +2353,100 @@ def test_integration_stop_reason(mock_client):
     assert props["$ai_stop_reason"] in ("stop", "length")
     assert props["$ai_provider"] == "openai"
     assert props["$ai_input_tokens"] > 0
+
+
+@pytest.mark.asyncio
+async def test_async_chat_streaming_supports_async_with(
+    mock_client, streaming_tool_call_chunks
+):
+    """Regression test for #393: chat completions stream=True must support
+    `async with` (the protocol pydantic-ai relies on)."""
+
+    async def mock_create(self, **kwargs):
+        return RecordingAsyncStream(streaming_tool_call_chunks)
+
+    with patch(
+        "openai.resources.chat.completions.AsyncCompletions.create", new=mock_create
+    ):
+        client = AsyncOpenAI(api_key="test-key", posthog_client=mock_client)
+
+        response = await client.chat.completions.create(
+            model="gpt-4",
+            messages=[{"role": "user", "content": "Hi"}],
+            stream=True,
+            posthog_distinct_id="test-id",
+        )
+
+        chunks = []
+        async with response as stream:
+            async for chunk in stream:
+                chunks.append(chunk)
+
+    assert chunks == streaming_tool_call_chunks
+    assert mock_client.capture.call_count == 1
+    call_args = mock_client.capture.call_args[1]
+    props = call_args["properties"]
+    assert call_args["event"] == "$ai_generation"
+    assert props["$ai_provider"] == "openai"
+    assert props["$ai_model"] == "gpt-4"
+
+
+@pytest.mark.asyncio
+async def test_async_responses_streaming_supports_async_with(mock_client):
+    """Regression test for #393: responses stream=True must support
+    `async with`."""
+    from unittest.mock import MagicMock
+
+    chunk = MagicMock()
+    chunk.type = "response.text.delta"
+    chunk.text = "hello"
+
+    async def mock_create(self, **kwargs):
+        return RecordingAsyncStream([chunk])
+
+    with patch("openai.resources.responses.AsyncResponses.create", new=mock_create):
+        client = AsyncOpenAI(api_key="test-key", posthog_client=mock_client)
+
+        response = await client.responses.create(
+            model="gpt-4o-mini",
+            input=[{"role": "user", "content": "Hi"}],
+            stream=True,
+            posthog_distinct_id="test-id",
+        )
+
+        async with response as stream:
+            received = [c async for c in stream]
+
+    assert received == [chunk]
+    assert mock_client.capture.call_count == 1
+
+
+@pytest.mark.asyncio
+async def test_async_chat_streaming_early_exit_closes_provider_stream(
+    mock_client, streaming_tool_call_chunks
+):
+    """Breaking out of the stream early must close the underlying provider
+    stream (release the HTTP connection) and still capture the event."""
+    source = RecordingAsyncStream(streaming_tool_call_chunks)
+
+    async def mock_create(self, **kwargs):
+        return source
+
+    with patch(
+        "openai.resources.chat.completions.AsyncCompletions.create", new=mock_create
+    ):
+        client = AsyncOpenAI(api_key="test-key", posthog_client=mock_client)
+
+        response = await client.chat.completions.create(
+            model="gpt-4",
+            messages=[{"role": "user", "content": "Hi"}],
+            stream=True,
+            posthog_distinct_id="test-id",
+        )
+
+        async with response as stream:
+            async for _ in stream:
+                break
+
+    assert source.closed is True
+    assert mock_client.capture.call_count == 1