Peer-to-Peer LAN
Clew Code instances can discover and coordinate with each other on the same machine or LAN.
Discovery
- Same machine — file-based peer registry
- LAN — UDP multicast discovery
- HTTP heartbeat — 60s liveness checks
Commands
| Command | Description |
|---|---|
/peer share | Advertise as worker on LAN |
/peer discover | Find other instances |
/peer join | Connect to a peer |
/peer send | Send message/task to peer |
/peer swarm | Run command on ALL peers |
/peer dashboard | Show peer task checklist |
/peer memory sync | Import memories from peers |
/peer memory auto on | Auto-sync every 60 min |
15+ AI tools
PeerBroadcastTool, PeerDashboardTool, PeerDiscoverTool, PeerInfoTool, PeerJoinTool, PeerListMessagesTool, PeerListRolesTool, PeerPingTool, PeerRunTool, PeerSendMessageTool, PeerSetNameTool, PeerSetRoleTool, PeerShareTool, PeerSpawnTool, PeerSwarmTool.
Swarm execution
Broadcasts a shell command to ALL connected peers in parallel with aggregated results. Supports --timeout, --filter, --dry-run.
Task persistence
The swarm task queue is persisted to disk (~/.clew/peer-swarm-tasks.json) so it survives restarts. Running tasks interrupted by a shutdown are rehydrated as failed with an "Interrupted by restart" error.
Worktree isolation
Each swarm task can run in its own isolated git worktree, preventing concurrent or queued tasks from interfering with files an interactive session (or another task) is using. Opt-in per server via isolateWorktrees option.
Task dependencies
Tasks can declare dependencies (dependsOn). A queued task is skipped until its depended-upon tasks complete. Task history is bounded so dependsOn resolution doesn't grow unbounded.
Concurrency
By default, peers execute one task at a time. The maxConcurrentTasks option controls how many tasks run in parallel on a given peer.
Message broker
In-process queue for offline message delivery with correlation IDs.
Peer memory sync
Import memories from all peers into local MemoryDB. Auto-sync available on cron schedule.