Orchestrator
The orchestrator is a background process that automates the entire work lifecycle: pulling tickets, spawning agents, monitoring progress, running reviews, and scaling your team.
What the Orchestrator Does
Instead of manually running prlt work spawn and monitoring each agent, the orchestrator handles it all:
- Watches the board — Detects tickets in the Ready column
- Spawns agents — Automatically assigns and launches work
- Monitors health — Restarts stuck or crashed agents
- Runs reviews — Triggers the review pipeline when agents finish
- Scales capacity — Matches active agents to available work
Starting the Orchestrator
# Start the orchestrator in a tmux session $ prlt orchestrator start # Check if it's running $ prlt orchestrator status # Attach to the orchestrator session $ prlt orchestrator attach # Stop the orchestrator $ prlt orchestrator stop
How It Works
The orchestrator runs as a long-lived process in a tmux session. It polls the board on a regular interval and takes action:
- Pull — Moves tickets from Backlog to Ready using diet ratios
- Spawn — Assigns agents to Ready tickets and starts sessions
- Monitor — Checks session health and restarts if needed
- Review — When agents finish, queues the review pipeline
- Complete — Moves reviewed tickets to Done
Watch Mode
An alternative to the full orchestrator, watch mode auto-spawns agents as tickets enter a specific column:
# Watch the Ready column and auto-spawn $ prlt work watch --column "Ready" # Watch with Docker isolation $ prlt work watch --column "Ready" --environment docker
Tip
Watch mode is lighter than the full orchestrator — it only handles spawning, not reviews or health monitoring. Use it for simpler workflows.
Caffeinate
On macOS, prevent your machine from sleeping while agents work:
# Keep macOS awake while agents run $ prlt caffeinate start # Check status $ prlt caffeinate status # Stop when done $ prlt caffeinate stop