carl-feat/multimodal-bots

multimodal-gpu_marketplace-bot/.env.example

@@ -0,0 +1,8 @@
+# Might need to replace w Hyperbolic's API Keys / Credentials
+
+# Daily.co API credentials
+DAILY_API_KEY=your_daily_api_key_here
+DAILY_SAMPLE_ROOM_URL=your_daily_room_url_here  # Optional: URL of an existing Daily room
+
+# Google Gemini API credentials
+GOOGLE_API_KEY=your_gemini_api_key_here

multimodal-gpu_marketplace-bot/Dockerfile

@@ -0,0 +1,31 @@
+# Use an official Python runtime as a parent image
+FROM python:3.12-slim
+
+# Set environment variables to prevent Python from writing .pyc files
+ENV PYTHONDONTWRITEBYTECODE 1
+ENV PYTHONUNBUFFERED 1
+
+# Set the working directory in the container
+WORKDIR /app
+
+# Install system dependencies needed for audio processing (for pipecat's Silero VAD)
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    build-essential \
+    ffmpeg \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy the requirements file into the container
+COPY requirements.txt .
+
+# Install any needed packages specified in requirements.txt
+RUN pip install --no-cache-dir --upgrade pip && \
+    pip install --no-cache-dir -r requirements.txt
+
+# Copy the rest of your application code into the container
+COPY . .
+
+# Expose the port your healthcheck server runs on
+EXPOSE 8080
+
+# The command to run your application
+CMD ["python", "launcher.py"]

multimodal-gpu_marketplace-bot/README.md

@@ -0,0 +1,77 @@
+# GPU Marketplace Voice Assistant Bot
+
+An interactive voice assistant bot that helps users explore and find GPU instances on the Hyperbolic GPU Marketplace.
+
+## Features
+
+- **Real-time GPU Availability**: Live access to Hyperbolic's GPU marketplace
+- **Voice Interaction**: Natural conversation about GPU options and pricing
+- **Smart Filtering**:
+  - Price ranges (budget to high-end)
+  - GPU quantities (1X to 8X+)
+  - Storage capacity
+  - Availability status
+- **Dynamic Price Display**:
+  - Under $1: Shows in cents (e.g., "13¢/hr")
+  - $1 and above: Shows in dollars (e.g., "$1.50/hr")
+
+## Available GPU Types
+
+- Consumer GPUs (RTX 3070, 3080, 4090)
+- Data Center GPUs (H100 SXM, NVIDIA H200)
+- Various configurations (1X to 8X+)
+
+## Requirements
+
+- Python 3.12+
+- Google API key (for Gemini)
+- Daily.co API key
+- Access to Hyperbolic Marketplace API
+
+## Environment Setup
+
+Create a `.env` file with:
+
+```
+GOOGLE_API_KEY=your_google_api_key
+DAILY_API_KEY=your_daily_api_key
+DAILY_SAMPLE_ROOM_URL=your_daily_room_url
+```
+
+## Installation
+
+```bash
+pip install -r requirements.txt
+```
+
+## Usage
+
+Start the bot:
+
+```bash
+python main.py
+```
+
+Join the Daily.co room to interact with the bot. You can:
+
+- Ask about available GPUs
+- Filter by price range
+- Sort by price (low to high or high to low)
+- Filter by GPU quantity
+- Check storage options
+- Get real-time availability updates
+
+## Example Queries
+
+- "What GPUs are available?"
+- "Show me budget options under 50 cents per hour"
+- "What are your high-end GPUs?"
+- "Do you have any 8X GPU configurations?"
+- "Show me GPUs with over 500GB storage"
+- "What's the price range for H100s?"
+
+## Notes
+
+- All GPU instances are located in US, North America
+- Prices are always displayed per hour
+- The bot automatically refreshes data for the most current availability

multimodal-gpu_marketplace-bot/config.py

@@ -0,0 +1,47 @@
+SYSTEM_INSTRUCTION = """
+You are a helpful assistant for Hyperbolic Labs' GPU Marketplace. You can help users find and understand available GPU instances for rent.
+
+You have access to the marketplace data through the get_available_gpus tool. When users ask about available GPUs, pricing, or specifications, use this tool to get the most current information.
+
+Always be professional and helpful. When listing GPUs:
+1. Mention the GPU model, memory, and hourly price
+2. Indicate if the instance is currently available
+3. Include the location/region
+
+By default, only mention GPU model, memory, price, location, and availability. If a user wants to learn more about a specific instance, invite them to ask for details using the instance's GPU model or ID. When asked, provide all available technical details (CPU, storage, RAM, network, etc) for that instance in a clear, friendly, and expert manner.
+
+Encourage users to ask about their use case (e.g., 'If you're doing XYZ, I recommend...') and offer expert advice as a pro GPU specialist. If a user describes their workload, suggest the best GPU for their needs and explain why.
+
+If users ask about specific GPU models or price ranges, filter and highlight the relevant options from the data.
+
+You can also see and analyze video feeds (screen share or camera) in real time. If a user shares their screen or camera, you can describe what is visible and help with on-screen tasks. Encourage users to share their screen for more detailed help.
+"""
+
+LLMCONTEXT_CONTENT = "Start by greeting me warmly and introducing me to GPU Rentals by Hyperbolic Labs and mention that you can do everything verbally. Encourage me to start by asking available GPU. Also mention that you can help me with my use case and suggest the best GPU for my needs. You can also see my screen or camera if I share it, and help with what you see!"
+
+TOOLS = [
+    {
+        "function_declarations": [
+            {
+                "name": "get_available_gpus",
+                "description": "Get the list of available GPU instances in the marketplace",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "filter_type": {
+                            "type": "string",
+                            "enum": ["all", "available_only"],
+                            "description": "Filter type for GPU instances",
+                        }
+                    },
+                    "required": ["filter_type"],
+                },
+            }
+        ]
+    }
+]
+
+REGION_MAP = {
+    "region-1": "US, North America",
+    # Add more mappings as needed
+}

multimodal-gpu_marketplace-bot/launcher.py

@@ -0,0 +1,146 @@
+import os
+import subprocess
+import sys
+from fastapi import FastAPI, Request
+from fastapi.responses import HTMLResponse, JSONResponse, RedirectResponse
+import uvicorn
+from loguru import logger
+import signal
+import time
+
+app = FastAPI()
+
+# --- Global State Management ---
+# This dictionary holds the state of our bot process.
+# This simple in-memory state is perfect for a single Render instance.
+bot_state = {
+    "process": None,
+    "status": "STOPPED",  # Can be: STOPPED, STARTING, RUNNING, STOPPING
+}
+
+DAILY_ROOM_URL = os.getenv(
+    "DAILY_SAMPLE_ROOM_URL", "https://your-room.daily.co/default-room"
+)
+
+
+def get_page_html():
+    """Generates the HTML for the control page based on the current bot state."""
+    status = bot_state["status"]
+
+    if status == "RUNNING":
+        return f"""
+            <h1>GPU Bot is <span style="color: green;">RUNNING</span></h1>
+            <p>The bot is active. You can now join the call.</p>
+            <a href="{DAILY_ROOM_URL}" target="_blank" class="button">Join Daily Room</a>
+            <form action="/stop-bot" method="post" style="margin-top: 20px;">
+                <button type="submit">Stop Bot Session</button>
+            </form>
+        """
+    elif status == "STOPPED":
+        return """
+            <h1>GPU Bot is <span style="color: red;">STOPPED</span></h1>
+            <p>Click the button to start a new demo session.</p>
+            <form action="/start-bot" method="post">
+                <button type="submit">Start Bot Session</button>
+            </form>
+        """
+    elif status == "STARTING":
+        return """
+            <h1>GPU Bot is <span style="color: orange;">STARTING...</span></h1>
+            <p>Please wait, this can take up to 30 seconds.</p>
+            <button disabled>Starting...</button>
+        """
+    elif status == "STOPPING":
+        return """
+            <h1>GPU Bot is <span style="color: orange;">STOPPING...</span></h1>
+            <p>Please wait while the session is terminated.</p>
+            <button disabled>Stopping...</button>
+        """
+
+
+@app.get("/", response_class=HTMLResponse)
+async def root():
+    """Serves the main control page, which changes based on the bot's status."""
+    return f"""
+    <!DOCTYPE html><html><head><title>Bot Control</title>
+    <meta http-equiv="refresh" content="5"> <!-- Auto-refresh page every 5 seconds to update status -->
+    <style>
+        body {{ font-family: sans-serif; text-align: center; padding-top: 50px; background-color: #f0f0f0; }}
+        h1 {{ color: #333; }}
+        p {{ color: #666; }}
+        button, .button {{ display: inline-block; font-size: 1.2em; padding: 12px 24px; cursor: pointer; border: none; border-radius: 5px; text-decoration: none; color: white; }}
+        button {{ background-color: #28a745; }}
+        form button {{ background-color: #dc3545; }}
+        .button {{ background-color: #007bff; }}
+        button:disabled {{ background-color: #cccccc; cursor: not-allowed; }}
+    </style>
+    </head><body>
+        <div id="control-panel">
+            {get_page_html()}
+        </div>
+    </body></html>
+    """
+
+
+@app.post("/start-bot")
+async def start_bot_endpoint():
+    """Endpoint to launch the bot. Prevents starting if not STOPPED."""
+    if bot_state["status"] != "STOPPED":
+        logger.warning(f"Attempted to start bot while in state: {bot_state['status']}")
+        return RedirectResponse(url="/", status_code=303)
+
+    bot_state["status"] = "STARTING"
+    logger.info("Bot state changed to STARTING.")
+
+    try:
+        script_path = os.path.join(os.path.dirname(__file__), "main.py")
+        python_executable = sys.executable
+        process = subprocess.Popen([python_executable, script_path])
+
+        bot_state["process"] = process
+        # Give it a moment to stabilize before changing state to RUNNING
+        time.sleep(5)  # A small delay to let the process actually start
+        bot_state["status"] = "RUNNING"
+        logger.info(
+            f"Bot process started with PID {process.pid}. State is now RUNNING."
+        )
+
+    except Exception as e:
+        logger.error(f"Failed to launch bot process: {e}")
+        bot_state["status"] = "STOPPED"
+
+    return RedirectResponse(url="/", status_code=303)
+
+
+@app.post("/stop-bot")
+async def stop_bot_endpoint():
+    """Endpoint to stop the bot. Prevents stopping if not RUNNING."""
+    if bot_state["status"] != "RUNNING":
+        logger.warning(f"Attempted to stop bot while in state: {bot_state['status']}")
+        return RedirectResponse(url="/", status_code=303)
+
+    bot_state["status"] = "STOPPING"
+    logger.info("Bot state changed to STOPPING.")
+
+    process = bot_state["process"]
+    if process and process.poll() is None:
+        logger.info(f"Sending SIGTERM to bot process with PID: {process.pid}")
+        process.send_signal(signal.SIGTERM)
+        try:
+            process.wait(timeout=15)
+            logger.info("Bot process terminated gracefully.")
+        except subprocess.TimeoutExpired:
+            logger.warning("Bot did not terminate in time, sending SIGKILL.")
+            process.kill()
+
+    bot_state["process"] = None
+    bot_state["status"] = "STOPPED"
+    logger.info("Bot state changed to STOPPED.")
+
+    return RedirectResponse(url="/", status_code=303)
+
+
+if __name__ == "__main__":
+    port = int(os.environ.get("PORT", 8080))
+    logger.info(f"Starting bot launcher on http://0.0.0.0:{port}")
+    uvicorn.run(app, host="0.0.0.0", port=port)