Running Headless Orchestrator

Overview

A headless orchestrator is an agent that:

1. Starts the relay infrastructure itself (agent-relay up)
2. Spawns and manages worker agents
3. Monitors agent lifecycle events
4. Coordinates work without human intervention

When to Use

• Agent needs full control over its worker team
• No human available to run agent-relay up manually
• Agent should manage agent lifecycle autonomously
• Building self-contained multi-agent systems

Quick Reference

Bootstrap Flow

Step 0: Verify Installation

# Check if agent-relay is available
command -v agent-relay || npx agent-relay --version

# If your shell reports a mise/asdf shim error, fix Node first
node --version
# e.g. for mise: mise use -g node@22.22.1

# If not installed, install globally
npm install -g agent-relay

# Or use npx (no global install)
npx agent-relay --version

Step 1: Start Infrastructure

# Preferred: run broker in foreground/stdin mode and keep the session open
agent-relay up --no-dashboard --verbose

Step 2: Spawn Workers via MCP

mcp__relaycast__agent_add(
  name: "Worker1",
  cli: "claude",
  task: "Implement the authentication module following the existing patterns"
)

Step 3: Monitor and Coordinate

# Check if workers have replied (returns unread counts — not the content)
mcp__relaycast__message_inbox_check()

# List Worker1's DM conversations (use `as` to specify the agent)
mcp__relaycast__message_dm_list(as: "Worker1")

# Send a targeted DM to a specific worker
mcp__relaycast__message_dm_send(to: "Worker1", text: "Also add unit tests")

# Broadcast to all agents on a channel
mcp__relaycast__message_post(channel: "general", text: "All workers: wrap up current task")

# List active workers
mcp__relaycast__agent_list()

Step 4: Release Workers

mcp__relaycast__agent_remove(name: "Worker1")

Step 5: Shutdown (optional)

agent-relay down

CLI Commands for Orchestration

Channel vs DM — When to Use Each

# WRONG — history (no flags) will not show DM replies from workers
agent-relay history

# Read a worker's UNREAD DM replies (clears after reading)
agent-relay inbox --agent Worker1

# Read the full DM conversation history with a worker (read + unread)
agent-relay history --to Worker1

# Read only the thread between two specific agents
agent-relay history --to Worker1 --from Orchestrator

Spawning and Messaging

# Spawn a worker
agent-relay spawn Worker1 claude "Implement auth module"

# Send a DM to a specific worker (replies readable via inbox --agent)
agent-relay send Worker1 "Add unit tests too"

# Broadcast to all workers via channel
agent-relay send '#general' "Team: wrap up and report status"

# Read Worker1's DM reply
agent-relay inbox --agent Worker1

# Release when done
agent-relay release Worker1

Monitoring Workers (Essential)

# Show currently active agents
agent-relay who

# View real-time output from a worker (critical for debugging)
agent-relay agents:logs Worker1

# Read DM replies from a specific worker
agent-relay inbox --agent Worker1

# View channel message history (channel posts only — not DMs)
agent-relay history --to '#general'

# Check overall system status
agent-relay status

Troubleshooting

# Kill unresponsive worker
agent-relay agents:kill Worker1

# Re-check broker status
agent-relay status

# If a worker looks stuck, inspect its logs first
agent-relay agents:logs Worker1

Orchestrator Instructions Template

Give your lead agent these instructions:

You are an autonomous orchestrator. Bootstrap the relay infrastructure and manage a team of workers.

## Step 1: Verify Installation
Run: command -v agent-relay || npx agent-relay --version
If you hit a mise/asdf shim error: verify Node first with `node --version`, then fix the runtime manager
If not found: npm install -g agent-relay

## Step 2: Start Infrastructure
Run: agent-relay up --no-dashboard --verbose
Verify: agent-relay status (should show "running")

## Step 3: Manage Your Team

Spawn workers:
  agent-relay spawn Worker1 claude "Task description"

Monitor workers (do this frequently):
  agent-relay who              # List active workers
  agent-relay agents:logs Worker1  # View worker output/progress

Send targeted DM instructions:
  agent-relay send Worker1 "Additional instructions"

Broadcast to all workers:
  agent-relay send '#general' "All workers: prioritize the auth module"

Read worker replies (DMs are not visible in history):
  agent-relay inbox --agent Worker1

Release when done:
  agent-relay release Worker1

## Protocol
- Workers will ACK when they receive tasks
- Workers will send DONE when complete
- Use `agent-relay agents:logs <name>` to monitor progress
- Use `agent-relay inbox --agent <name>` to read **unread** DM replies from a worker (clears after reading)
- Use `agent-relay history --to <name>` to re-read the full DM conversation (read + unread)
- Use `agent-relay history --to '#general'` to see channel message flow
- Do NOT use `agent-relay history` alone to check worker replies — it only shows channel posts, DM replies are invisible there

Lifecycle Events

The broker emits these events (available via SDK subscriptions):

Common Mistakes

Overview

Self-bootstrap agent-relay infrastructure and manage a team of agents autonomously.

Prerequisites

1. agent-relay CLI installed (required)

npm install -g agent-relay
   # Or use npx without installing: npx agent-relay <command>