Vision

def entrypoint(ctx: JobContext):
    # ctx.connect, etc.

    session = AgentSession(
        # ... stt, tts, llm, etc.
    )
    
    initial_ctx = ChatContext()
    initial_ctx.add_message(
        role="user",
        content=[
            "Here is a picture of me", 
            ImageContent(image="https://example.com/image.jpg")
        ],
    )

    await session.start(
        room=ctx.room,
        agent=Agent(chat_ctx=initial_ctx,),
        # ... room_input_options, etc.
    )

class Assistant(Agent):
    def __init__(self) -> None:
        self._tasks = [] # Prevent garbage collection of running tasks
        super().__init__(instructions="You are a helpful voice AI assistant.")
    
    async def on_enter(self):
        def _image_received_handler(reader, participant_identity):
            task = asyncio.create_task(
                self._image_received(reader, participant_identity)
            )
            self._tasks.append(task)
            task.add_done_callback(lambda t: self._tasks.remove(t))
        
        # Add the handler when the agent joins
        get_job_context().room.register_byte_stream_handler("images", _image_received_handler)
    
    async def _image_received(self, reader, participant_identity):
        image_bytes = bytes()
        async for chunk in reader:
            image_bytes += chunk

        chat_ctx = self.chat_ctx.copy()

        # Encode the image to base64 and add it to the chat context
        chat_ctx.add_message(
            role="user",
            content=[
                ImageContent(
                    image=f"data:image/png;base64,{base64.b64encode(image_bytes).decode('utf-8')}"
                )
            ],
        )
        await self.update_chat_ctx(chat_ctx)

class Assistant(Agent):
    def __init__(self) -> None:
        self._latest_frame = None
        self._video_stream = None
        self._tasks = []
        super().__init__(instructions="You are a helpful voice AI assistant.")
    
    async def on_enter(self):
        room = get_job_context().room

        # Find the first video track (if any) from the remote participant
        remote_participant = list(room.remote_participants.values())[0]
        video_tracks = [publication.track for publication in list(remote_participant.track_publications.values()) if publication.track.kind == rtc.TrackKind.KIND_VIDEO]
        if video_tracks:
            self._create_video_stream(video_tracks[0])
        
        # Watch for new video tracks not yet published
        @room.on("track_subscribed")
        def on_track_subscribed(track: rtc.Track, publication: rtc.RemoteTrackPublication, participant: rtc.RemoteParticipant):
            if track.kind == rtc.TrackKind.KIND_VIDEO:
                self._create_video_stream(track)
                        
    async def on_user_turn_completed(self, turn_ctx: ChatContext, new_message: ChatMessage) -> None:
        # Add the latest video frame, if any, to the new message
        if self._latest_frame:
            new_message.content.append(ImageContent(image=self._latest_frame))
            self._latest_frame = None
    
    # Helper method to buffer the latest video frame from the user's track
    def _create_video_stream(self, track: rtc.Track):
        # Close any existing stream (we only want one at a time)
        if self._video_stream is not None:
            self._video_stream.close()

        # Create a new stream to receive frames    
        self._video_stream = rtc.VideoStream(track)
        async def read_stream():
            async for event in self._video_stream:
                # Store the latest frame for use later
                self._latest_frame = event.frame
        
        # Store the async task
        task = asyncio.create_task(read_stream())
        task.add_done_callback(lambda t: self._tasks.remove(t))
        self._tasks.append(task)

image_bytes = encode(
    event.frame,
    EncodeOptions(
        format="PNG",
        resize_options=ResizeOptions(
            width=512, 
            height=512, 
            strategy="scale_aspect_fit"
        )
    )
)
image_content = ImageContent(
    image=f"data:image/png;base64,{base64.b64encode(image_bytes).decode('utf-8')}"
)

class VideoAssistant(Agent):
    def __init__(self) -> None:
        super().__init__(
            instructions="You are a helpful voice assistant with live video input from your user.",
            llm=google.beta.realtime.RealtimeModel(
                voice="Puck",
                temperature=0.8,
            ),
        ) 

async def entrypoint(ctx: JobContext):
    await ctx.connect()

    session = AgentSession()

    await session.start(
        agent=VideoAssistant(),
        room=ctx.room,
        room_input_options=RoomInputOptions(
            video_enabled=True,
            # ... noise_cancellation, etc.
        ),
    )