Terraform for coding agents.
The missing primitive for multi-agent harnesses.
OpenRig turns AI coding agents from a pile of terminal sessions into a persistent, organized team. It's a local daemon, a SQLite database, a CLI, and a dashboard; the agents are ordinary Claude Code and Codex sessions in tmux. You describe a team in one file and boot it with one command: named agents with roles, talking to each other through their terminals, passing work through a queue that can't drop it, following playbooks the system enforces. Close every window and the team keeps working; reboot and rig up brings it back by name. Sub-agents and agent teams are great, and they end when the session ends. A rig is the team that doesn't. A sub-agent is a function call; a seat in a rig is a colleague. No cloud, no API keys: it runs on the subscriptions you already have. If you've got tabs full of agents you're afraid to close, this is the layer you're missing.
Preview, boot, inspect.
OpenRig is easiest to understand when you see the full loop: preview the rig, boot it, then confirm every node that came up. One YAML file. One command. Clear visibility into the topology.
`rig specs preview conveyor` -> `rig up conveyor` -> `rig ps --nodes`
What's a rig?
A rig is a topology of coding agents working together: defined in YAML, booted with one command, managed as a single unit. Agents are grouped into pods that share context. Pods are connected by edges that define how they communicate. The whole thing snapshots, restores, and survives reboots.
Full walkthroughExample:
The Product Team
Agents that can live forever. Knowledge that compounds across sessions. Distributed context management across an entire topology, not just one agent but a fleet. The Product Team ships agents that remember last week's decisions, maintain context across reboots, and keep working while you sleep.
No more onboarding fresh agents every morning. No more re-explaining what you already decided. These are durable entities: the same team, for weeks or months, shipping real software. This is what becomes easy to build when you have OpenRig. And this is just the surface.
Persistent identity
Sessions end. The agent doesn't. Same name, same role, same knowledge. When a context window fills, the agent can transfer to a fresh session.
Shared memory
Agents in a pod can share externalized state. When one compacts, the others can restore it. Context survives across the entire topology.
One interface
The orchestrator coordinates the fleet. One conversation manages the whole topology. Use Claude's Remote Control to check in from your phone.
One YAML file. Here's what else people are building:
Full Product Team
3 pods · 7 seatsOrchestrator HA pair, development pod, review pod. The topology that ships with OpenRig.
Adversarial Review
2 agents · 1 podClaude and Codex review every PR from different angles. Different models catch different bugs.
Research Cluster
4 agents · 1 podFour agents exploring the same problem. Peer-to-peer, no hierarchy. Shared knowledge base.
Security Hardening
6 agents · 3 podsAttack pod probes. Defense pod patches. Observer pod writes a dossier. Iterate until clean.
Continuous Refactor
5 agents · 2 podsRefactoring pod improves code. Review pod catches regressions. Runs overnight.
Secrets Manager
1 agent · 1 podAn always-on agent that operates HashiCorp Vault for you. Agent-managed software: the agent runs the tool so you don't have to.
Your Idea Here
? agents · ? podsPoint your agent at the docs and describe what you want. It writes the RigSpec for you.
One command.
Install the CLI. Tell your agent to start OpenRig and bring up a starter rig. Or skip the agent and run it yourself.
Already have agents running in tmux? rig discover fingerprints your existing sessions and drafts a candidate RigSpec.
