Integration
Meshcore
By MeshCore KI Integration — LoRa mesh + AI gateway. Dual AI: Claude Sonnet (remote) + Gemma 4 (local failover). Off-grid chat, IoT control, 133-char chunking. Standalone Python + OpenClaw plugin (planned).
Configuration Example
meshcore:
port: "auto" # Serial port or "auto" for auto-detection
baud: 115200
ai:
ollama_url: "http://127.0.0.1:11434"
ollama_model: "gemma3:4b"
anthropic_model: "claude-sonnet-4-20250514"
prefer_local: false # Set true to prefer local Gemma over remote Claude
timeout: 30
enable_iot: true
iot:
enabled: true
dry_run: true # Set false when GPIO hardware is connected
access:
dm_policy: "open" # "open" or "allowlist"
require_prefix: "" # e.g. "!ai " to only respond to prefixed messages
privacy:
store_messages: false
redact_logs: true
disable_context: false
context_timeout_minutes: 10README
# MeshCore KI Integration
**AI-powered LoRa mesh gateway** — bridges [MeshCore](https://github.com/rpsreal/MeshCore) mesh networks with AI models for off-grid intelligent communication.
Send a text message from any MeshCore LoRa node → the gateway processes it through AI → you get an answer back over radio. No internet needed at the endpoint.
> **Two editions available:**
> - **Standalone** — slim Python scripts, direct Anthropic API, zero dependencies beyond an API key
> - **OpenClaw Edition** *(planned)* — MeshCore as a channel plugin for [OpenClaw](https://openclaw.ai), the open-source AI gateway
---
## Features
- **Dual AI Backend** — Claude Sonnet (remote, via Anthropic API) + Gemma 4 (local, via Ollama). Automatic failover between online and offline AI.
- **Off-Grid AI Chat** — Ask anything over LoRa radio. Hazmat identification, first aid protocols, weather, technical references — all without mobile coverage.
- **IoT Control** — Natural language commands like "turn on basecamp lights" get translated into GPIO actions on the gateway.
- **Message Chunking** — AI responses are automatically split into 133-character chunks (MeshCore limit) with `[n/N]` prefixes.
- **Privacy First** — No messages stored on disk. Log redaction enabled by default. Conversation context lives in RAM only, auto-purged after configurable timeout.
- **Access Control** — Open or allowlist-based DM policy, blocklist, optional message prefix gating.
- **Compact & Reliable** — Failover depends only on the API key. No cron jobs, no complex setup.
## Architecture
### Standalone Edition
```
┌───────────────┐ LoRa 868 MHz ┌─────────────────────┐
│ MeshCore Node│ ◄──────────────────► │ WIO Pro P1 Tracker │
│ (Handheld/App)│ MeshCore Protocol │ (USB Gateway Radio) │
└───────────────┘ └─────────┬───────────┘
│
USB Serial │ 115200 Baud
│
┌──────────────┴──────────────┐
│ Server (Debian Linux) │
│ │
│ meshcore_client.py │
│ meshcore_service.py │
│ ai_backend.py │
│ iot_handler.py │
└──────┬──────────┬───────────┘
│ │
▼ ▼
┌───────────────┐ ┌─────────────────┐
│ Claude Sonnet │ │ Gemma 4 │
│(Anthropic API)│ │ (Ollama) │
│ Remote/Primary│ │ Local/Fallback │
└───────────────┘ └─────────────────┘
```
### OpenClaw Edition *(planned)*
```
┌───────────────┐ ┌────────────────────┐
│ LoRa Mesh │ │ Chat Apps │
│ (MeshCore) │ │ (WhatsApp, TG...) │
└───────┬───────┘ └────────┬───────────┘
│ │
▼ ▼
┌───────────────────────────────────────────────────────────┐
│ OpenClaw Gateway │
│ │
│ MeshCore Channel Plugin │ Chat Plugins │ AI Router │
│ Memory Manager │ Context Engine │
└──────────────┬────────────────────────┬───────────────────┘
│ │
▼ ▼
┌───────────────┐ ┌─────────────────┐
│ Claude Sonnet │ │ Gemma 4 │
│ (Primary) │ │ (Fallback) │
└───────────────┘ └─────────────────┘
```
## Hardware
| Component | Details |
|-----------|---------|
| Gateway Radio | [Seeed WIO Pro P1 Tracker](https://www.seeedstudio.com/) |
| Interface | USB Serial, 115200 Baud, appears as `/dev/ttyACM0` |
| Server | Any Linux machine (Debian, Ubuntu, Raspberry Pi) |
| LoRa Nodes | Any MeshCore-compatible device |
| Cost | ~€50 per node |
## Quick Start
```bash
# Clone the repository
git clone https://github.com/gunpowder100/openclaw-meshcore.git
cd openclaw-meshcore
# Run setup script (creates venv, installs deps, detects hardware)
chmod +x setup.sh
./setup.sh
# Set your Anthropic API key (optional, for remote AI)
export ANTHROPIC_API_KEY="sk-ant-..."
# Run
source .venv/bin/activate
python openclaw_meshcore.py
```
## Configuration
Edit `config.yaml` to customize:
```yaml
meshcore:
port: "auto" # Serial port or "auto" for auto-detection
baud: 115200
ai:
ollama_url: "http://127.0.0.1:11434"
ollama_model: "gemma3:4b"
anthropic_model: "claude-sonnet-4-20250514"
prefer_local: false # Set true to prefer local Gemma over remote Claude
timeout: 30
enable_iot: true
iot:
enabled: true
dry_run: true # Set false when GPIO hardware is connected
access:
dm_policy: "open" # "open" or "allowlist"
require_prefix: "" # e.g. "!ai " to only respond to prefixed messages
privacy:
store_messages: false
redact_logs: true
disable_context: false
context_timeout_minutes: 10
```
## Dual AI — Why Two Models?
| | Claude Sonnet (Primary) | Gemma 4 (Fallback) |
|---|---|---|
| **Location** | Remote (Anthropic API) | Local (Ollama) |
| **Internet** | Required | Not needed |
| **Knowledge** | Full reasoning, large context | Limited by storage & CPU |
| **Use case** | Complex queries, analysis | Basic Q&A, offline mode |
| **Trigger** | Default | Auto-failover when API unreachable |
> **Note:** Gemma 4 runs with limited knowledge scope due to server storage and CPU constraints — ideal for basic queries and predefined topics. The failover is fully automatic and depends only on the API key availability. No cron jobs or complex setup required.
## Mesh Commands
| Command | Description |
|---------|-------------|
| `!ai <text>` | Query the AI (if prefix is configured) |
| `!status` | Show system status |
| `!ping` | Test connectivity |
## Test Environment
| Component | Details |
|-----------|---------|
| Server | Hermes — Debian Linux, 192.168.178.73 |
| Gateway | WIO Pro P1 Tracker via USB (/dev/ttyACM0) |
| AI Primary | Claude Sonnet via Anthropic API (requires ANTHROPIC_API_KEY) |
| AI Fallback | Gemma 4 (gemma3:4b) via Ollama on localhost:11434 |
| LoRa Band | 868 MHz (EU) |
| Python | 3.13+ with asyncio |
| Packages | meshcore, httpx, pyyaml |
## Roadmap
- [x] Standalone Python integration
- [x] Dual AI failover (Claude + Gemma)
- [x] Channel message support
- [x] IoT/GPIO command parsing
- [ ] OpenClaw channel plugin integration
- [ ] Multi-channel bridging (LoRa ↔ WhatsApp/Telegram)
- [ ] Web dashboard
## License
MIT — see [LICENSE](LICENSE)
## Author
**Jonathan Salim** — Buergerfunkinitiative Essen-Kettwig
- GitHub: [gunpowder100](https://github.com/gunpowder100)
- Web: [jonathansalim.de](https://jonathansalim.de)
- Email: [email protected]
Copyright © 2026 Jonathan Salim. All rights reserved.
integration
Comments
Sign in to leave a comment