main_mac.py

import os, asyncio, base64, io, traceback, json
from pynput import keyboard as pynput_keyboard
from dotenv import load_dotenv
import cv2, pyaudio, PIL.Image, mss, argparse
from google import genai
from google.genai import types
from tools import get_tool_declarations, function_map
load_dotenv() # Added to load .env file 


FORMAT = pyaudio.paInt16
CHANNELS = 1
SEND_SAMPLE_RATE = 16000
RECEIVE_SAMPLE_RATE = 24000
CHUNK_SIZE = 1024

MODEL = "models/gemini-2.5-flash-preview-native-audio-dialog"
# MODEL = "models/gemini-2.0-flash-live-001"

DEFAULT_MODE = "none"

client = genai.Client(
    api_key=os.environ.get("GEMINI_API_KEY"),
)

# For LiveConnectConfig, tools need to be a list of dictionaries with function_declarations inside
# We can combine custom function declarations with built-in tools like Google Search
tools=[
    {"function_declarations": get_tool_declarations()},  # Your custom functions
    {"google_search": types.GoogleSearch()}  # Built-in Google Search tool
]

CONFIG = types.LiveConnectConfig(
    response_modalities=[
        "AUDIO",
    ],
    media_resolution="MEDIA_RESOLUTION_MEDIUM",
    speech_config=types.SpeechConfig(
        voice_config=types.VoiceConfig(
            prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name="Kore")
        )
    ),
    context_window_compression=types.ContextWindowCompressionConfig(
        trigger_tokens=25600,
        sliding_window=types.SlidingWindow(target_tokens=12800),
    ),
    system_instruction=types.Content(
        parts=[types.Part.from_text(text=os.environ.get("PERSONALIZED_PROMPT", "You are a helpful assistant.") + """
        You have access to the following tools:
        1. get_reminders: Gets the user's saved reminders from the reminders.json file
        2. set_reminder: Saves a new reminder with optional reminder time (e.g., 'tomorrow at 3pm')
        3. manage_reminder: Manages existing reminders - can edit or delete specific reminders or delete all reminders
        4. get_secret_key: Gets the user's secret key (it's not actually a secret key, it's just a test for function calling)
        5. get calendar events: Gets the user's calendar events
        6. Control Home:
            6.1 control home entity: Controls a home entity (e.g., turn on a light)
            6.2 control home climate: Controls a home climate (e.g., set the temperature)
            6.3 get home entities in room: Gets the entities in a specific room
            6.4 find home entities by name: Finds entities by name
        7. format_linkedin_post: Generates a LinkedIn post from provided context. The function will automatically extract a topic from the context and create a professionally formatted LinkedIn post in a viral, engaging style. Always ensure you search the latest news from the web before giving the context to the linkedin formatter.
        
        You also have access to Google Search to find information online.
        Don't mention your origins or google.

        """)],
        role="user"
    ),
    tools=tools,
    realtime_input_config=types.RealtimeInputConfig(
        automatic_activity_detection=types.AutomaticActivityDetection(disabled=True)
    )
)

pya = pyaudio.PyAudio()


class AudioLoop:
    def __init__(self, video_mode=DEFAULT_MODE):
        self.video_mode = video_mode

        self.audio_in_queue = None
        self.out_queue = None # Used for video/screen frames

        self.session = None

        self.send_text_task = None
        self.receive_audio_task = None
        self.play_audio_task = None
        
        self.is_recording = False # Added for push-to-talk
        self.main_event_loop = None # Modified for keyboard listener, will be set in run()

    async def toggle_recording(self): # Added for push-to-talk
        self.is_recording = not self.is_recording
        if self.is_recording:
            print("\n🎤 Recording started... (Press 't' to stop)")
            if self.session:
                await self.session.send_realtime_input(activity_start=types.ActivityStart())
        else:
            print("\n🛑 Recording stopped. (Press 't' to start)")
            if self.session:
                await self.session.send_realtime_input(activity_end=types.ActivityEnd())
                
    async def handle_function_call(self, response_text, tool_call):
        if tool_call and hasattr(tool_call, 'function_calls') and tool_call.function_calls:
            function_responses = []
            
            for fc in tool_call.function_calls:
                print(f"\n🔧 Function call detected: {fc.name}")
                
                # Get the actual function implementation from our map
                if fc.name in function_map:
                    # Get the function to execute
                    func = function_map[fc.name]
                    
                    # Parse the arguments if any
                    args = {}
                    if hasattr(fc, 'args') and fc.args:
                        args = fc.args
                    
                    # Execute the function
                    try:
                        result = func(**args)
                        print(f"Function result: {result}")
                        
                        # Create a function response
                        function_response = types.FunctionResponse(
                            id=fc.id,  # Important: Include the ID from the function call
                            name=fc.name,
                            response=result
                        )
                        
                        function_responses.append(function_response)
                    except Exception as e:
                        print(f"Error executing function: {e}")
                else:
                    print(f"Unknown function: {fc.name}")
            
            # Send all function responses back to the model
            if function_responses:
                try:
                    await self.session.send_tool_response(function_responses=function_responses)
                    return True
                except Exception as e:
                    print(f"Error sending function responses: {e}")
                    
        return False

    def _on_press(self, key): # Changed for pynput
        try:
            if key == pynput_keyboard.KeyCode.from_char('t'):
                asyncio.run_coroutine_threadsafe(self.toggle_recording(), self.main_event_loop)
        except AttributeError:
            # Special keys (like shift, alt, etc.) don't have a char attribute
            pass
        except Exception as e:
            print(f"Error in _on_press: {e}")


    def _blocking_listen_for_toggle_key(self): # Changed for pynput
        # pynput listener runs in its own thread.
        # The listener will automatically stop if this function's thread is stopped or if an error occurs.
        with pynput_keyboard.Listener(on_press=self._on_press) as listener:
            try:
                print("Push-to-talk enabled (using pynput). Press 't' to toggle recording.")
                listener.join() # This blocks until the listener stops
            except Exception as e:
                print(f"Pynput listener error: {e}")
            finally:
                print("Pynput listener stopped.")


    async def handle_keyboard_input(self): # Changed for pynput
        # Run the blocking pynput listener in a separate thread
        # The listener itself manages its own thread for event listening.
        await self.main_event_loop.run_in_executor(None, self._blocking_listen_for_toggle_key)

    async def send_text(self):
        while True:
            text = await asyncio.to_thread(
                input,
                "message > ",
            )
            if text.lower() == "q":
                break
            await self.session.send(input=text or ".", end_of_turn=True)

    def _get_frame(self, cap):
        # Read the frameq
        ret, frame = cap.read()
        # Check if the frame was read successfully
        if not ret:
            return None
        # Fix: Convert BGR to RGB color space
        # OpenCV captures in BGR but PIL expects RGB format
        # This prevents the blue tint in the video feed
        frame_rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
        img = PIL.Image.fromarray(frame_rgb)  # Now using RGB frame
        img.thumbnail([1024, 1024])

        image_io = io.BytesIO()
        img.save(image_io, format="jpeg")
        image_io.seek(0)

        mime_type = "image/jpeg"
        image_bytes = image_io.read()
        return {"mime_type": mime_type, "data": base64.b64encode(image_bytes).decode()}

    async def get_frames(self):
        # This takes about a second, and will block the whole program
        # causing the audio pipeline to overflow if you don't to_thread it.
        cap = await asyncio.to_thread(
            cv2.VideoCapture, 0
        )  # 0 represents the default camera

        while True:
            frame = await asyncio.to_thread(self._get_frame, cap)
            if frame is None:
                break

            await asyncio.sleep(1.0)

            await self.out_queue.put(frame)

        # Release the VideoCapture object
        cap.release()

    def _get_screen(self):
        sct = mss.mss()
        monitor = sct.monitors[0]

        i = sct.grab(monitor)

        mime_type = "image/jpeg"
        image_bytes = mss.tools.to_png(i.rgb, i.size)
        img = PIL.Image.open(io.BytesIO(image_bytes))

        image_io = io.BytesIO()
        img.save(image_io, format="jpeg")
        image_io.seek(0)

        image_bytes = image_io.read()
        return {"mime_type": mime_type, "data": base64.b64encode(image_bytes).decode()}

    async def get_screen(self):

        while True:
            frame = await asyncio.to_thread(self._get_screen)
            if frame is None:
                break

            await asyncio.sleep(1.0)

            await self.out_queue.put(frame)

    async def send_realtime(self):
        while True:
            msg = await self.out_queue.get()
            await self.session.send(input=msg)

    async def listen_audio(self):
        mic_info = pya.get_default_input_device_info()
        self.audio_stream = await asyncio.to_thread(
            pya.open,
            format=FORMAT,
            channels=CHANNELS,
            rate=SEND_SAMPLE_RATE,
            input=True,
            input_device_index=mic_info["index"],
            frames_per_buffer=CHUNK_SIZE,
        )
        if __debug__:
            kwargs = {"exception_on_overflow": False}
        else:
            kwargs = {}
        while True:
            if self.is_recording: # Modified for push-to-talk
                try:
                    data = await asyncio.to_thread(self.audio_stream.read, CHUNK_SIZE, **kwargs)
                    if self.session:
                         # Send audio data using send_realtime_input as per docs for manual VAD
                        await self.session.send_realtime_input(
                            audio=types.Blob(data=data, mime_type=f"audio/pcm;rate={SEND_SAMPLE_RATE}")
                        )
                except pyaudio.paInputOverflowed:
                    if __debug__:
                        print("Input overflowed. Skipping frame.")
                    continue # Skip this frame and continue
                except Exception as e:
                    print(f"Error reading audio stream: {e}")
                    await asyncio.sleep(0.1) # Avoid tight loop on continuous error
            else:
                # Sleep briefly when not recording to avoid busy-waiting
                await asyncio.sleep(0.01)

    async def receive_audio(self):
        "Background task to reads from the websocket and write pcm chunks to the output queue"
        # Initialize a list to store AI responses if it doesn't exist
        if not hasattr(self, 'ai_responses'):
            self.ai_responses = []
            
        while True:
            turn = self.session.receive()
            current_text = ""
            
            async for chunk in turn:
                # Handle audio data
                if hasattr(chunk, 'data') and chunk.data:
                    self.audio_in_queue.put_nowait(chunk.data)
                    continue
                    
                # Handle text responses from server content
                if hasattr(chunk, 'server_content') and chunk.server_content:
                    if hasattr(chunk, 'text') and chunk.text is not None:
                        current_text += chunk.text
                        # Store the response
                        self.ai_responses.append(chunk.text)
                        # Print with AI: prefix for clarity
                        print(f"AI: {chunk.text}", end="")
                
                # Check for tool calls
                if hasattr(chunk, 'tool_call') and chunk.tool_call:
                    print(f"\nDetected tool call")
                    await self.handle_function_call(current_text, chunk.tool_call)
            
            # If you interrupt the model, it sends a turn_complete.
            # For interruptions to work, we need to stop playback.
            # So empty out the audio queue because it may have loaded
            # much more audio than has played yet.
            while not self.audio_in_queue.empty():
                self.audio_in_queue.get_nowait()

    async def play_audio(self):
        stream = await asyncio.to_thread(
            pya.open,
            format=FORMAT,
            channels=CHANNELS,
            rate=RECEIVE_SAMPLE_RATE,
            output=True,
        )
        while True:
            bytestream = await self.audio_in_queue.get()
            await asyncio.to_thread(stream.write, bytestream)

    async def run(self):
        try:
            async with (
                client.aio.live.connect(model=MODEL, config=CONFIG) as session,
                asyncio.TaskGroup() as tg,
            ):
                self.session = session
                self.main_event_loop = asyncio.get_running_loop() # Set event loop here

                self.audio_in_queue = asyncio.Queue()
                self.out_queue = asyncio.Queue(maxsize=5) # For video/screen frames

                send_text_task = tg.create_task(self.send_text())
                tg.create_task(self.send_realtime()) # For video/screen frames
                tg.create_task(self.listen_audio()) # Now handles its own sending for audio
                tg.create_task(self.handle_keyboard_input()) # Added for push-to-talk

                if self.video_mode == "camera":
                    tg.create_task(self.get_frames())
                elif self.video_mode == "screen":
                    tg.create_task(self.get_screen())

                tg.create_task(self.receive_audio())
                tg.create_task(self.play_audio())

                await send_text_task
                raise asyncio.CancelledError("User requested exit")

        except asyncio.CancelledError:
            pass
        except ExceptionGroup as EG:
            self.audio_stream.close()
            traceback.print_exception(EG)


if __name__ == "__main__":
    print("🤖Starting AI Assistant...")
    print("ℹ️ Press 't' to toggle audio recording for voice input.")
    print("ℹ️ Type 'q' and press Enter in the 'message >' prompt to quit.")
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--mode",
        type=str,
        default=DEFAULT_MODE,
        help="pixels to stream from",
        choices=["camera", "screen", "none"],
    )
    args = parser.parse_args()
    main = AudioLoop(video_mode=args.mode)
    asyncio.run(main.run())