AI Responder

The AI Responder is an optional sidecar container that transforms your Meshtastic node into an AI-powered assistant. Users on the mesh can ask questions, have conversations, and get intelligent responses — all through standard Meshtastic messaging.

Credit

The AI Responder was created by LN4CY. MeshMonitor integrates it as an optional Docker sidecar.

Overview

The AI Responder connects to MeshMonitor's Virtual Node (port 4404) and monitors configured channels for messages prefixed with !ai. It supports multiple AI providers — from local models via Ollama for privacy-focused deployments, to cloud providers like Google Gemini, OpenAI, and Anthropic for more powerful responses.

Key features:

• Multi-provider support: Ollama (local), Google Gemini, OpenAI, Anthropic Claude
• Per-user conversation history: Each user gets isolated conversation context
• Session mode: Continuous conversations in DMs without repeating the !ai prefix
• Mesh-optimized: Responses tuned for LoRa message size limits (~200 characters)
• Automatic reconnection: Resilient connection with 10-second retry loop

How It Works

┌─────────────────────────────────────────────────────────┐
│                     Your Server                         │
│                                                         │
│  ┌────────────────┐     ┌──────────────┐                │
│  │  MeshMonitor   │────►│ AI Responder │                │
│  │  (port 3001)   │     │  (sidecar)   │                │
│  │                │     │              │                │
│  │  Virtual Node  │◄────│  TCP Client  │                │
│  │  (port 4404)   │     │              │───► AI Provider│
│  └────────┬───────┘     └──────────────┘    (Ollama,   │
│           │                                  Gemini,   │
│           │                                  OpenAI,   │
│           │                                  Claude)   │
└───────────┼─────────────────────────────────────────────┘
            │
            │ TCP (port 4403)
            ▼
    ┌───────────────┐
    │  Meshtastic   │
    │    Node       │
    └───────────────┘

Key Points:

1. The AI Responder connects to MeshMonitor's Virtual Node (port 4404)
2. It monitors configured channels for !ai messages
3. Questions are forwarded to the configured AI provider
4. Responses are sent back through the mesh network

Quick Start

Prerequisites

1. Virtual Node enabled — The AI Responder requires MeshMonitor's Virtual Node feature (ENABLE_VIRTUAL_NODE=true)
2. AI provider — Either a local Ollama instance or a cloud API key (Gemini, OpenAI, or Anthropic)

Docker Compose Setup

Add the AI Responder to your existing docker-compose.yml:

services:
  meshmonitor:
    image: ghcr.io/yeraze/meshmonitor:latest
    # ... your existing MeshMonitor configuration ...
    environment:
      - ENABLE_VIRTUAL_NODE=true
      - VIRTUAL_NODE_PORT=4404
      # ... other environment variables ...

  # AI Responder - AI-powered mesh assistant
  # Credit: https://github.com/LN4CY/ai-responder
  ai-responder:
    image: ghcr.io/ln4cy/ai-responder:latest
    container_name: meshmonitor-ai-responder
    restart: unless-stopped
    environment:
      - INTERFACE_TYPE=tcp
      - MESHTASTIC_HOST=meshmonitor
      - MESHTASTIC_PORT=4404
      - AI_PROVIDER=ollama           # or: gemini, openai, anthropic
      - OLLAMA_HOST=ollama           # if using Ollama
      - ADMIN_NODE_ID=!your_admin_id # your node ID for admin commands
      - ALLOWED_CHANNELS=0           # comma-separated channel indices
    volumes:
      - ai-data:/app/data
    depends_on:
      - meshmonitor

  # Optional: Local AI with Ollama
  ollama:
    image: ollama/ollama:latest
    container_name: meshmonitor-ollama
    restart: unless-stopped
    volumes:
      - ollama-data:/root/.ollama

volumes:
  ai-data:
  ollama-data:

Deploy and Initialize

# Start the stack
docker compose up -d

# If using Ollama, pull a model
docker exec -it meshmonitor-ollama ollama pull llama3.2:1b

Configuration Options

The AI Responder is configured via environment variables:

Variable	Default	Description
INTERFACE_TYPE	tcp	Connection type: tcp or serial
MESHTASTIC_HOST	meshmonitor	Hostname of MeshMonitor (use Docker service name)
MESHTASTIC_PORT	4404	Virtual Node port
AI_PROVIDER	ollama	AI provider: ollama, gemini, openai, or anthropic
OLLAMA_HOST	ollama	Ollama service hostname (when using Ollama)
GEMINI_API_KEY	-	Google Gemini API key
OPENAI_API_KEY	-	OpenAI API key
ANTHROPIC_API_KEY	-	Anthropic API key
ADMIN_NODE_ID	-	Node ID with admin privileges (e.g., !1234abcd)
ALLOWED_CHANNELS	0,3	Comma-separated list of channel indices to monitor
HISTORY_MAX_MESSAGES	1000	Maximum stored messages per user
HISTORY_MAX_BYTES	2097152	Maximum history file size (bytes)
CHUNK_DELAY	15	Seconds between message chunks
CONNECTION_RETRY_INTERVAL	10	Seconds between reconnection attempts

Supported AI Providers

Provider	Type	Best For
Ollama	Local	Privacy-focused deployments, no internet required
Google Gemini	Cloud	Powerful responses with optional grounding (search/maps)
OpenAI	Cloud	GPT-powered responses
Anthropic	Cloud	Claude-powered responses

Admins can switch providers at runtime using the !ai -p command.

User Commands

Basic Usage

Command	Description
!ai <question>	Ask a question (required in channels, one-off in DMs)
!ai -h	Show help menu

Session Management (DM Only)

Command	Description
!ai -n [name]	Start a continuous conversation session
!ai -end	End the active session

Sessions allow back-and-forth conversation in DMs without repeating the !ai prefix. They auto-timeout after 5 minutes of inactivity.

Conversation History

Command	Description
!ai -c ls	List saved conversations (max 10 slots per user)
!ai -c [name/number]	Load a previous conversation
!ai -c rm [name/number]	Delete a specific conversation
!ai -c rm all	Clear all saved conversations
!ai -m	Show context usage and storage stats

Admin Commands

Command	Description
!ai -p	List available AI providers
!ai -p [provider]	Switch AI provider (admin only)

Channel vs. DM Behavior

Feature	Channel	DM
!ai prefix required	Every message	First message only (in session)
Session mode	Not available	Available via !ai -n
Conversation history	Per-channel, per-user	Per-user
Slot usage	Does not count	Uses 1 of 10 slots

Troubleshooting

Checking Logs

docker compose logs ai-responder -f

Common Issues

"Connection refused" or "Cannot connect"

• Ensure MeshMonitor is running and Virtual Node is enabled
• Verify MESHTASTIC_HOST matches your MeshMonitor service name
• Check that MESHTASTIC_PORT matches your Virtual Node port (default: 4404)

No response to !ai messages

• Verify the channel index is in ALLOWED_CHANNELS
• Check that the AI provider is configured correctly
• If using Ollama, ensure a model has been pulled (docker exec ollama ollama list)

Slow or truncated responses

• LoRa messages are limited in size; responses are automatically chunked
• Adjust CHUNK_DELAY if responses are being dropped
• Cloud providers generally give faster responses than local Ollama

Provider API errors

• Verify your API key is correct for the selected provider
• Check that the provider service is reachable from the container
• Review logs for specific error messages

Related Documentation

• Virtual Node - Required for AI Responder
• MQTT Client Proxy - Another community add-on by LN4CY
• Community Add-ons Overview - All available add-ons
• LN4CY ai-responder Repository - Full documentation and source code