Skip to content

Latest commit

 

History

History
141 lines (97 loc) · 4.76 KB

README.md

File metadata and controls

141 lines (97 loc) · 4.76 KB
Raw

LLM/Bot Platform with Meshtastic

This repository provides a platform for integrating Large Language Models (LLMs) or Old-school Bots with the Meshtastic mesh communication network. It allows users on the mesh network to interact with an LLM for concise, automated responses.

Furthermore, this platform also allows users to execute tasks through LLM, such as:

  • Calling emergency services
  • Sending messages
  • Retrieving info from sensors

Note: Currently the only supported tool is a demo for emergency service. More tools are coming soon. You can add your own tool using the guide below

Features

  • Bi-directional communication between Meshtastic and an LLM.
  • Support for general broadcast or targeted responses.
  • Automatic message chunking for long responses exceeding 200 characters.
  • Maintains message history for context-aware interactions.
  • Node-specific information (e.g., battery level, location) can be included in responses.
  • Tool Use: Your LLM can execute tasks for you based on your prompt
  • Old-school bots features: anything you can imagine with a basic telegram or discord bots.

Requirements

  • Python 3.8+
  • Meshtastic Python library
  • Ollama LLM Python SDK
  • PubSub library

Setup

  1. Connect your Meshtastic device via USB or configure it for TCP access.
  2. Clone this repository: 
    git clone <repo_url>
cd <repo_name>
  3. Install dependencies: 
    pip install -r requirement.txt
  4. Run the script: 
    python main.py
  5. To talk with the LLM, you can use normal message or "/tool your_message" to activate tool use.

Ollama performace with Tool Use on small model (llama 3.2:3b) is not exactly correct. Please test your model extensively before putting it into real use.

How to use

Radio LLM works like a discord chat bot. You can send commands into the channel to configure chat or run commands.

For LLM chat features
/enable_llm
/disable_llm
For echo features to test meshtastic
/enable_echo
/disable_echo

Configuration

  • Modify the LLM model by updating the chat function in chat_with_llm.
  • Adjust chunk size or message length limits as needed.

Custom Tools

To add your own tool:

  1. Define your tool in model/tool_handler.py
  2. Register your tool in model/tool_registry.py
  3. Describe your tool in config.yaml

Please use the same name for your tool across all steps. In the future, this process will be streamlined.

Different Interface

If you use BLE on your computer, please check Meshtastic documentation here first. It will help you navigate the meshtastic cli to search for devices and how to authenticate the connection.

# Use this if your node is connected to your local network
interface = meshtastic.tcp_interface.TCPInterface(hostname="meshtastic.local")

# Use this if your node is on BLE
# Before using BLE client, you should connect to your device using your system bluetooth settings.
# Read more on https://meshtastic.org/docs/software/python/cli/usage/#utilizing-ble-via-the-python-cli
interface = meshtastic.ble_interface.BLEClient(address="Your Node BLE Identifier")

# Use this if your node is connected to your computer
interface = meshtastic.serial_interface.SerialInterface() # add param devPath if you have multiple devices connected

Ollama Model

If you use Ollama, please change the model name in model/config.yaml to your installed model.

How It Works

  1. Receiving Messages:

    • The script listens for incoming messages on the Meshtastic network.
    • Received messages trigger the LLM to generate a response.

  2. Generating Responses:

    • The chat_with_llm function interacts with the LLM using the ollama library.
    • Responses are concise and limited to 200 characters.

  3. Sending Responses:

    • Responses are sent back to the sender or broadcasted to the network.
    • Messages exceeding 200 characters are sent in chunks.

  4. Node-Specific Information:

  • The script can retrieve and include specific details about the sending node, such as:
    • Node ID
    • Battery level
    • Location (latitude, longitude, altitude)
    • Last heard time
  • This information can be appended to responses for context-aware conversations.

Key Components

  • onReceive(packet, interface): Handles incoming messages.
  • chat_with_llm(user_id, message): Queries the LLM and returns a response.
  • onConnection(interface): Manages connection to the Meshtastic device.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contributions

Feel free to submit issues or pull requests for improvements or bug fixes.

Disclaimer

Ensure compliance with local laws and regulations when using Meshtastic devices and LLMs.