Most multi-agent experiments end up as a set of fragile terminal tabs: lost context, hard-to-recover sessions, and no simple way to evolve the team. OpenRig treats a working group of coding agents as an operable system you can declare, boot, inspect, snapshot and restore — so long-running, specialist seats accumulate context and continue across reboots instead of disappearing into terminal sprawl.
What Sets It Apart
- Declarative topologies (RigSpec): define pods, seats and edges in YAML so the team layout is versionable and portable; rigs can be bundled and installed on another machine.
- Local control plane with visible primitives: a small daemon + SQLite database, a CLI (
rig) and a TUI let you bootrig up, stoprig down, inspect seats, and attach to agent tmux sessions — you operate the team on your machine rather than using a cloud service. - Mixed harness support and seat continuity: runs Claude Code, Codex and Pi sessions as ordinary tmux seats, discovers and adopts existing sessions, and snapshots state so conversations and owned work can be resumed or audited.
- Workflow and safety primitives: per-seat permission modes, activity relays, managed hooks and workspace trust rules let operators balance automation with human oversight while keeping workspace writes and hooks auditable.
Who it's for — and tradeoffs
Great fit if you want a reproducible, self-hosted way to run LLM-driven coding teams that need persistent context (builders, reviewers, QA) and you prefer direct access to terminals and the underlying processes. It is also useful for teams experimenting with agent topologies, bundles and agent-managed services (examples include a Vault specialist or service-backed rigs). Look elsewhere if you need a turnkey cloud-managed agent platform, must run natively on Windows (native Windows isn't supported), or cannot meet the requirements (Node.js 22/24, tmux, macOS/Linux). OpenRig coordinates existing provider-backed agents — it does not replace the models or provider billing, and some setup writes provider hooks and workspace trust files that operators should review before first use.
