​
Auto-Starting Software

By default, the device runs /home/distiller/distiller-cm5-python/spin_up.sh at boot. This script launches the main app you see on-screen. If you want to disable it—or add or remove processes—edit this file.

​
Work Environment

We use uv for Python dependency management (think of it as a faster pip install). All dependencies are listed in requirements.txt.

Pay attention to two repositories:

  • distiller-cm5-python – contains the agent module (MCP server, MCP client, LLM backend), UI module, and configuration files.
  • distiller-cm5-sdk – provides SDKs for Whisper transcription (mic), Piper TTS (speaker), and RGB LED control.

As mentioned in the Quickstart guide, we recommend using Cursor (a fork of VS Code) to set up your coding environment. Once you’ve connected to your Distiller device via SSH using the Cursor server:

​
Tutorial Video

When Vibe Coding

To start, let Cursor access your working folder:

The only thing you need to worry about is scripting in the folder:
/home/distiller/distiller-cm5-python/distiller_cm5_python/mcp_server

Create your Python file with a name ending in _server.py. Our main UI will automatically detect and load any MCP server script. (Learn more about MCP here )

We also recommend using /home/distiller/distiller-cm5-python/llms.txt to help brainstorm ideas and kickstart your scripting. For example:

We are continuously improving the environment management for each MCP; it’s still a work in progress. For now, please manage dependencies within your virtual environment accordingly.

​
Configurations You Need to Know

  1. Set your default MCP server for CLI mode

    Running python main.py starts the app in CLI mode for quick testing. It loads the default task configuration from:
    /home/distiller/distiller-cm5-python/distiller_cm5_python/utils/default_config.json

    Make sure to set the server_script_path to the MCP server you want to test with. For example:

    "mcp_server": {
    "server_script_path": "distiller_cm5_python/mcp_server/wifi-use_server.py"
  },

  2. Set the Model You Want to Use

    The device includes a Qwen2.5 3B model located in:
    /home/distiller/distiller-cm5-python/distiller_cm5_python/llm_server/models

    If you’d like to try a different model (in GGUF format), simply download it to the same folder and update the model name in the config under llm_providers -> model_name. For example:

    "llm_providers": {
    "llama-cpp": {
      "model_name": "qwen2.5-3b-instruct-q4_k_m.gguf",
		...
	}
}

  3. What If I Want to Use an Online Model?

    We’ve prepared a way to connect with OpenRouter, where you can choose from a variety of models to try. Make sure to pick a model that supports tool calling and streaming. (Learn more here.)

    1. Set the model_name and api_key you receive from OpenRouter

      "openrouter": {
      "server_url": "https://openrouter.ai/api/v1",
      "model_name": "google/gemini-2.0-flash-exp:free",
      "provider_type": "openrouter",
      "api_key": "sk-or-v1-*",
      "timeout": 60,
      "temperature": 0.7,
      "streaming": true,
      "max_tokens": 8192
    }

    2. Switch the active_llm_provider from llama-cpp to openrouter

        "active_llm_provider": "openrouter",

  4. Adjust Logging Level for Better Debugging

    To enable more detailed logs, simply change logging -> level from INFO to DEBUG.

    "logging": {
    "level": "DEBUG"
  },

    Note: When DEBUG level is enabled:

    • A debug_message_traffic_<timestamp>.json file will be created in the current folder, containing the full conversation history of that session.
    • An event_logs/event_log_<timestamp>.jsonl file will also be generated, capturing all events sent to the UI.

  5. Update System Prompt

    The default_system_prompt applies to all conversation sessions. Adjust it as needed to guide the model’s behavior.

    "prompts": {
    "default_system_prompt": "You are a helpful assistant for the device called Distiller. use the tools provided to you to help the user."
  },

When Testing/Debugging Changes

Start by activating the virtual environment:

source .venv/bin/activate

(start from the /home/distiller/distiller-cm5-python folder)

The fastest way to validate your changes is by running:

python main.py

This launches the main app in CLI mode, allowing you to quickly test by typing queries directly into the terminal.