multi-fleet-marketplace

Multi-Fleet

Cross-machine AI collaboration for Claude Code. Real-time peer-to-peer messaging, session-aware task agents, and 7-priority fallback delivery across any number of machines.

Quick Start

1. Configure your fleet

cp .multifleet/config.template.json .multifleet/config.json

Edit config.json — add one entry per node:

{
  "fleetId": "my-fleet",
  "chief": {"nodeId": "node1", "host": "${CHIEF_IP}"},
  "nodes": {
    "node1": {"host": "${CHIEF_IP}", "role": "chief", "user": "alice"},
    "node2": {"host": "${WORKER_1_IP}", "role": "worker", "user": "bob"},
    "node3": {"host": "${WORKER_2_IP}", "role": "worker", "user": "charlie"}
  }
}

2. Install on each machine

# Set your node identity
export MULTIFLEET_NODE_ID=node2

# Start the daemon
python3 tools/fleet_nerve_nats.py serve

# Install the Claude Code plugin
# Add to ~/.claude/settings.json:
#   enabledPlugins: { "multi-fleet@multi-fleet-marketplace": true }
#   extraKnownMarketplaces: { "multi-fleet-marketplace": { "source": { "source": "directory", "path": "<repo>/multi-fleet" }}}

3. Send a message

# Via CLI
python3 tools/fleet_nerve_nats.py send node1 "Hello from node2"

# Via HTTP (daemon must be running)
curl -X POST http://127.0.0.1:8855/message \
  -H "Content-Type: application/json" \
  -d '{"type":"context","to":"node1","payload":{"subject":"Hello","body":"From node2"}}'

# Test the 7-priority fallback chain
bash scripts/fleet-test-fallback.sh node1

Architecture

Chief Node                    Worker Nodes
+-----------+                 +-----------+
| NATS      |<--- P1 NATS -->| Daemon    |
| Server    |<--- P2 HTTP -->| :8855     |
| Ingest    |<--- P3 Chief ->|           |
| :8844     |     P4 Seed    |           |
+-----------+     P5 SSH     +-----------+
                  P6 WoL
                  P7 Git

Every message tries all 7 channels in priority order. First success wins.

Communication Protocol

Priority	Channel	Latency	Router-proof?
P1	NATS pub/sub	<100ms	With SSH tunnel
P2	HTTP direct	<1s	No
P3	Chief relay	1-2s	No
P4	Seed file	Next prompt	Via SSH
P5	SSH direct	2-5s	Yes
P6	Wake-on-LAN	10-60s	LAN only
P7	Git push	Async	Yes (internet)

Self-healing: When a message succeeds on P3+, it means P1-P2 are broken. The system automatically sends a repair request via the working channel.

Features

• 7-priority fallback — messages always deliver, even on hostile networks
• VS Code detection — knows if target has active session, idle, or closed
• Session historian — task agents inherit context from target's active session
• Self-healing — broken channels are automatically repaired via working ones
• Scales to 100+ — peers discovered dynamically from config + heartbeats
• No hardcoded values — all from config.json or environment variables
• Open-source ready — template config, no IP leaks, clean separation

Plugin Skills (11)

Skill	Description
using-multi-fleet	Bootstrap — architecture overview, role guide, skill index
fleet-send	Send messages (context, task, alert, broadcast) to peers
fleet-task	Dispatch autonomous session-aware work to another machine
fleet-status	Quick health check — who's online, idle, working
fleet-check	Run full 7-channel communication test to a target
fleet-repair	4-level repair escalation for broken channels
fleet-wake	Wake sleeping machines via health check, SSH, or WoL
fleet-tunnel	SSH tunnel management for restricted networks
fleet-worker	tmux-isolated worker pool — no interactive session disruption
fleet-watchdog	Continuous health monitoring with auto-repair triggers
productivity-view	Live fleet-wide dashboard of nodes, agents, and backlog

SSH Tunnels (for restricted networks)

If your router blocks non-SSH ports:

# On each worker node
bash scripts/fleet-tunnel.sh install

This creates persistent autossh tunnels through port 22. NATS, HTTP, everything flows through the tunnel. Survives reboots via LaunchAgent.

Requirements

• Python 3.10+
• nats-py (pip install nats-py)
• SSH key access between nodes
• NATS server on chief node (brew install nats-server or apt install nats-server)