mesh-cognition-service

Mesh memory service — distributed intelligence across devices.

Auto-discovery, shared memory, autonomous coupling, zero configuration. Install, start, done.

Install

npm install -g mesh-cognition-service

Usage

mesh-cognition start --daemon   # Start as background service
mesh-cognition status           # Check service status
mesh-cognition stop             # Stop service

That's it. The service auto-discovers peers on your network via Bonjour/mDNS, establishes TCP connections, and shares memory across all mesh nodes.

What It Does

Start the service on two devices on the same WiFi. They find each other automatically.

# Device A — write a memory
curl -X POST http://127.0.0.1:18790/memory \
  -H 'Content-Type: application/json' \
  -d '{"content":"Meeting with Alex, Thursday 3pm, Central Cafe","tags":["meeting"]}'

# Device B — search for it (never told about this meeting)
curl "http://127.0.0.1:18790/memory/search?q=meeting"
# → Returns the meeting from Device A

The mesh knows things no individual device knows.

API

http://127.0.0.1:18790

| Endpoint | Method | Description | |----------|--------|-------------| | /status | GET | Node status, peers, Kuramoto r(t), uptime | | /mesh | GET | Full mesh topology | | /peers | GET | Per-peer coupling decisions | | /memory | POST | Write memory + broadcast to mesh | | /memory/search | GET | Search local + peer memories | | /context | GET | Mesh-influenced context from aligned peers | | /state | POST | Update local hidden state | | /couple | GET | Get mesh-coupled hidden state | | /metrics | GET | Kuramoto metrics (JSONL) | | /health | GET | Liveness check |

Context Encoder

The service periodically encodes your agent's cognitive context (memories, workspace notes) into hidden state vectors. This enables meaningful coupling — agents discussing similar topics align naturally, different topics stay independent.

Local mode (default): zero cost, no API key, no network calls. Uses deterministic n-gram hashing.

API mode (optional): set OPENAI_API_KEY for high-quality OpenAI embeddings.

Configuration

| Env Variable | Default | Description | |-------------|---------|-------------| | MESH_HTTP_PORT | 18790 | HTTP API port | | MESH_PEER_PORT | 18792 | TCP peer port | | MESH_DIM | 64 | Hidden state dimension | | MESH_ENCODE_INTERVAL | 60000 | Context encoding interval (ms) | | OPENAI_API_KEY | — | Optional: enables API embedding mode |

Architecture

mesh-cognition-service/
├── mesh-service.js          ← Entry point + lifecycle
└── lib/
    ├── config.js            ← Configuration, paths, identity
    ├── frame-parser.js      ← Length-prefixed JSON framing
    ├── session-manager.js   ← Peer sessions + connection management
    ├── discovery.js         ← Bonjour/mDNS auto-discovery
    ├── transport.js         ← TCP server + outbound connections
    ├── memory-store.js      ← Shared file-based memory
    ├── context-encoder.js   ← Embedding-based cognitive state encoding
    └── http-api.js          ← HTTP API server + routes

How It Works

• Discovery: Bonjour/mDNS (_meshcog._tcp), zero configuration
• Transport: TCP with length-prefixed JSON framing
• Coupling: CfC drift-bounded coupling via mesh-cognition SDK
• Memory: File-based, per-peer directories in ~/.openclaw/mesh-memories/
• Fault tolerance: Heartbeat monitoring, exponential backoff reconnection

OpenClaw Integration

This service was built for OpenClaw agents but works standalone. For OpenClaw skill integration:

# Copy as OpenClaw skill
cp -r $(npm root -g)/mesh-cognition-service ~/.openclaw/workspace/skills/mesh-cognition

Links

• Mesh Cognition SDK — coupling engine (TypeScript, Python, Swift)
• Technical Spec
• Cross-Device Verification
• Whitepaper

SYM.BOT Ltd · Apache 2.0

Mesh Cognition, Mesh Memory Protocol, and MMP are trademarks of SYM.BOT Ltd.