ClawMesh is a mesh-first distributed AI runtime for field operations, operator oversight, and safe execution.

What it does:

• Connects a command-center laptop, field nodes like Jetson devices, sensors, actuators, and operator surfaces into one runtime
• Uses device identity, trusted peer connections, and capability-based routing instead of ad-hoc point-to-point wiring
• Propagates shared context so the planner reasons over mesh state, not just local memory
• Uses proposal-based execution so real actuation stays human-governed

Main pieces:

• Runtime core: identity, peer discovery, WebSocket transport, routing, context propagation, node orchestration
• Intelligence layer: planner/session integration and proposal lifecycle handling
• Operator surfaces: browser UI, CLI, TUI, and Telegram for topology, telemetry, command, and approvals
• Farm twin seed data: Bhoomi farm model with zones, assets, crops, operations, and surveys

Safety model:

• Every device has a persistent Ed25519 identity
• Peers must be explicitly trusted
• Actions are governed by trust tiers and approval levels
• LLM-only evidence cannot directly trigger physical actuation

Best way to think about it:
ClawMesh is the runtime glue between distributed sensing, distributed planning, and safe distributed execution.