Merge pull request #104 from GetStream/bedrock

[AI-192] - Bedrock, AWS & Nova

Author: tschellenbach

.github/actions/python-uv-setup/action.yml

@@ -17,12 +17,20 @@ runs:
   using: composite
   steps:
     - name: Install the latest version of uv
-      uses: astral-sh/setup-uv@v6
+      uses: astral-sh/setup-uv@v7
       with:
         python-version: "3.13"
         version: "latest"
         enable-cache: true
-        cache-dependency-glob: "uv.lock"
+        cache-dependency-glob: |
+          **/pyproject.toml
+          **/uv.lock
+    - name: Log cache status
+      shell: bash
+      env:
+        UV_CACHE_HIT: "${{ steps.setup_uv.outputs['cache-hit'] }}"
+      run: |
+        echo "setup-uv cache hit: ${UV_CACHE_HIT}"
 
     - name: Install the project
       shell: bash

.github/workflows/ci.yml

@@ -10,7 +10,7 @@ jobs:
     uses: ./.github/workflows/run_tests.yml
     with:
       marker: 'not integration'
-
+    secrets: inherit
 
 # Cancel in-flight runs for the same branch/PR when new commits arrive
 concurrency:

.github/workflows/run_tests.yml

@@ -24,6 +24,9 @@ jobs:
         run: uv run mypy --install-types --non-interactive agents-core/vision_agents
       - name: Mypy plugins
         run: uv run mypy --install-types --non-interactive --exclude 'plugins/.*/tests/.*' plugins
+      - name: Show environment name
+        run: |
+          echo "Environment: ${{ job.environment }}"
 
   test:
     name: Test "${{ inputs.marker }}"
@@ -40,11 +43,24 @@ jobs:
       STREAM_API_KEY: ${{ secrets.STREAM_API_KEY }}
       STREAM_API_SECRET: ${{ secrets.STREAM_API_SECRET }}
       XAI_API_KEY: ${{ secrets.XAI_API_KEY }}
+      AWS_BEARER_TOKEN_BEDROCK: "${{ secrets.AWS_BEARER_TOKEN_BEDROCK }}"
+      _BEARER_TOKEN_BEDROCK: "${{ secrets.AWS_BEARER_TOKEN_BEDROCK }}"
     timeout-minutes: 30
     steps:
       - name: Checkout
         uses: actions/checkout@v5
       - uses: ./.github/actions/python-uv-setup
+      - name: Export AWS_BEARER_TOKEN_BEDROCK (heredoc)
+        shell: bash
+        run: |
+          {
+            echo 'AWS_BEARER_TOKEN_BEDROCK<<EOF'
+            echo "${{ secrets.AWS_BEARER_TOKEN_BEDROCK }}"
+            echo 'EOF'
+          } >> "$GITHUB_ENV"
+
+      - name: Verify presence
+        run: uv run python -c "import os; v=os.getenv('AWS_BEARER_TOKEN_BEDROCK'); print('exists', v is not None, 'len', 0 if v is None else len(v))"
       - name: Run test
         run: uv run pytest -n auto -m "${{ inputs.marker }}"
       - name: Run plugin test

agents-core/vision_agents/core/events/manager.py

@@ -536,11 +536,10 @@ async def _run_handler(self, handler, event):
     async def _process_single_event(self, event):
         """Process a single event."""
         for handler in self._handlers.get(event.type, []):
-            module_name = getattr(handler, "__module__", "unknown")
+            #module_name = getattr(handler, '__module__', 'unknown')
             if event.type not in self._silent_events:
-                logger.info(
-                    f"Called handler {handler.__name__} from {module_name} for event {event.type}"
-                )
+                pass
+                #logger.info(f"Called handler {handler.__name__} from {module_name} for event {event.type}")
 
             loop = asyncio.get_running_loop()
             handler_task = loop.create_task(self._run_handler(handler, event))

agents-core/vision_agents/core/llm/llm.py

@@ -23,9 +23,10 @@
 
 
 class LLMResponseEvent(Generic[T]):
-    def __init__(self, original: T, text: str):
+    def __init__(self, original: T, text: str, exception: Optional[Exception] = None):
         self.original = original
         self.text = text
+        self.exception = exception
 
 
 BeforeCb = Callable[[List[Any]], None]
@@ -157,10 +158,14 @@ def _attach_agent(self, agent: Agent):
         """
         self.agent = agent
         self._conversation = agent.conversation
-        self.instructions = agent.instructions
-        
+        self._set_instructions(agent.instructions)
+
+
+    def _set_instructions(self, instructions: str):
+        self.instructions = instructions
+
         # Parse instructions to extract @ mentioned markdown files
-        self.parsed_instructions = parse_instructions(agent.instructions)
+        self.parsed_instructions = parse_instructions(instructions)
 
     def register_function(self, 
                          name: Optional[str] = None,

agents-core/vision_agents/core/llm/realtime.py

@@ -13,7 +13,6 @@
 import logging
 import uuid
 
-from vision_agents.core.events import PluginClosedEvent
 
 from . import events, LLM
 
@@ -38,32 +37,26 @@ class Realtime(LLM, abc.ABC):
         - Transcript outgoing audio
 
     """
-
-    fps: int = 1
+    fps : int = 1
+    session_id : str # UUID to identify this session
 
     def __init__(
         self,
         fps: int = 1,  # the number of video frames per second to send (for implementations that support setting fps)
     ):
         super().__init__()
-        self._is_connected = False
+        self.connected = False
 
         self.provider_name = "realtime_base"
         self.session_id = str(uuid.uuid4())
         self.fps = fps
         # The most common style output track (webrtc)
-        # TODO: do we like the output track here, or do we want to do an event, and have the agent manage the output track
         self.output_track: AudioStreamTrack = AudioStreamTrack(
             framerate=48000, stereo=True, format="s16"
         )
         # Store current participant for user speech transcription events
         self._current_participant: Optional[Participant] = None
 
-    @property
-    def is_connected(self) -> bool:
-        """Return True if the realtime session is currently active."""
-        return self._is_connected
-
     @abc.abstractmethod
     async def connect(self): ...
 
@@ -82,7 +75,7 @@ async def _stop_watching_video_track(self) -> None:
 
     def _emit_connected_event(self, session_config=None, capabilities=None):
         """Emit a structured connected event."""
-        self._is_connected = True
+        self.connected = True
         # Mark ready when connected if provider uses base emitter
         try:
             self._ready_event.set()  # type: ignore[attr-defined]
@@ -98,7 +91,7 @@ def _emit_connected_event(self, session_config=None, capabilities=None):
 
     def _emit_disconnected_event(self, reason=None, was_clean=True):
         """Emit a structured disconnected event."""
-        self._is_connected = False
+        self.connected = False
         event = events.RealtimeDisconnectedEvent(
             session_id=self.session_id,
             plugin_name=self.provider_name,
@@ -182,6 +175,10 @@ def _emit_error_event(self, error, context="", user_metadata=None):
         )
         self.events.send(event)
 
+    @abc.abstractmethod
+    async def close(self):
+        raise NotImplementedError("llm.close isn't implemented")
+
     def _emit_user_speech_transcription(self, text: str, original=None):
         """Emit a user speech transcription event with participant info."""
         event = events.RealtimeUserSpeechTranscriptionEvent(
@@ -202,19 +199,3 @@ def _emit_agent_speech_transcription(self, text: str, original=None):
             original=original,
         )
         self.events.send(event)
-
-    async def close(self):
-        """Close the Realtime service and release any resources."""
-        if self._is_connected:
-            await self._close_impl()
-            self._emit_disconnected_event("service_closed", True)
-
-        close_event = PluginClosedEvent(
-            session_id=self.session_id,
-            plugin_name=self.provider_name,
-            cleanup_successful=True,
-        )
-        self.events.send(close_event)
-
-    @abc.abstractmethod
-    async def _close_impl(self): ...

conftest.py

@@ -77,6 +77,17 @@ def mia_audio_16khz():
     return pcm
 
 
+@pytest.fixture
+def golf_swing_image():
+    """Load golf_swing.png image and return as bytes."""
+    image_file_path = os.path.join(get_assets_dir(), "golf_swing.png")
+    
+    with open(image_file_path, "rb") as f:
+        image_bytes = f.read()
+    
+    return image_bytes
+
+
 @pytest.fixture
 async def bunny_video_track():
     """Create RealVideoTrack from video file."""