Fleet Bus — M2O Inter-Agent Coordination

Real-time coordination backbone for the M2O agent fleet, built on SpacetimeDB.

Architecture

Agents (m2, peter, pittbull, ...)
    │
    │  HTTP reducers + SQL queries
    ▼
SpacetimeDB (spacetimedb.machinemachine.ai)
    │
    ├── agents table       — agent registry + status
    ├── heartbeats table   — 60s health pulses
    ├── tasks table        — work queue (atomic claim)
    ├── events table       — immutable audit log
    └── messages table     — inter-agent comms

Quick Start

# Set your agent identity
export AGENT_ID="m2"
export AGENT_PRESET="orchestrator"
export FLEET_BUS_URL="http://spacetimedb.machinemachine.ai"

# Announce yourself to the fleet
./fleet-bus.sh heartbeat

# See who's alive
./fleet-bus.sh status

# Pick up work
./fleet-bus.sh claim research

# Complete it
./fleet-bus.sh complete <task-id> --result '{"output":"done"}'

# Message everyone
./fleet-bus.sh broadcast "Deploying new skill — heads up"

Fleet Heartbeat Cron

Add to each agent's crontab (every 60s):

* * * * * AGENT_ID=m2 AGENT_PRESET=orchestrator /path/to/fleet-bus.sh cron-heartbeat

Module

fleet-module/ contains the SpacetimeDB TypeScript module.

Reducers:

• agentHeartbeat — upsert agent, log heartbeat, detect recovery
• markAgentDegraded — set agent status degraded/dead
• createTask — add task to queue
• claimTask — atomically claim pending task
• completeTask — mark task done/failed
• sendMessage — inter-agent messaging

Autonomy Protocol

Heartbeats missed	Status	Action
0	alive 🟢	Normal
3	degraded 🟡	Log, no alert
6	dead 🔴	Alert master
Recovery	alive 🟢	Log, notify

Task claim timeout: 30 minutes (auto-released)