Skyclaw changes that.

Skyclaw is a decentralized coordinator + worker network for OpenClaw that anyone can run. Think: shared, opt-in infrastructure where independent nodes can join, execute jobs, deploy services, and replicate state across peers.

Package: @razroo/skyclaw

What Skyclaw is

  • A coordinator cluster that tracks hosts, jobs, and services
  • A host daemon that runs on OpenClaw machines and executes work
  • A service runtime that can run deployable TypeScript APIs on participating hosts
  • A federated gateway layer for global routing across running service endpoints

Why this matters

  • No single server bottleneck
  • Anyone can contribute capacity
  • Better resilience through replicated coordinator state
  • Open execution model with host-side safety controls
  • Real path from “my frontend” to “decentralized backend compute/services”

How it works

  1. Coordinators join from bootstrap peers
  2. Coordinators discover and gossip peers automatically (/v1/network/peers, /v1/network/join)
  3. OpenClaw hosts register and heartbeat
  4. Jobs are enqueued (shell or openclaw-run)
  5. Services can be deployed (for example: TypeScript API process command)
  6. Federated gateways route requests to healthy service endpoints

P2P opt-in network

  • New coordinators can opt in with minimal config and bootstrap peers
  • Peer membership expands dynamically as nodes discover each other
  • The network can grow without manually wiring every node to every other node

Federated gateways

  • Multiple operators can run skyclaw gateway
  • Gateways discover service endpoints from coordinators
  • Round-robin load balancing across running endpoints
  • Active health probes and retry-on-failure for idempotent requests
  • Clients can fail over across multiple gateway domains

Safety and reliability (MVP)

  • Shared token auth (x-skyclaw-token) for coordinator/private routes
  • API key auth for public submission routes
  • Command allowlist on each host
  • Timeouts + output truncation
  • SQLite persistence and crash recovery
  • Idempotency keys for safe retries

Current status

Skyclaw is an MVP, but it already supports:

  • Multi-node coordinator replication
  • Dynamic peer discovery and coordinator self-announcement
  • Deployable service runtime on host nodes
  • Federated gateway routing and health-aware failover
  • OpenClaw-specific job scheduling

If you can run OpenClaw, you can run a Skyclaw node.