OpenRig, an Apache-2.0 project distributed as the npm package @openrig/cli, takes on the mess that appears when a team runs several coding agents at once. Rather than managing one agent, it manages the system they form: a 'rig' is a declared topology of pods, seats and edges, written in YAML and booted with a single command.
The distinctive choice is mixing vendors in one team. A rig can hold native Claude Code and Codex sessions side by side, addressed by stable names such as dev-owner@first-project, so an owner seat can hand a candidate change to a checker seat from the other harness. Under the hood it is a local daemon with an HTTP API, a CLI, a terminal UI and an MCP server, built on tmux and SQLite.
The project ships starting points rather than leaving topology design to the user: first-project (two seats), conveyor (a four-seat intake-plan-build-review handoff), product-team (two orchestrators plus implementation, QA, design and reviewers), adversarial-review, research-team and secrets-manager, the last running a HashiCorp Vault instance operated by a specialist agent. Rigs can be snapshotted with rig down --snapshot and restored by name, and existing tmux sessions can be fingerprinted and adopted into management.
The caveats are substantial and the documentation states them plainly. It requires Node.js 22 or 24 and tmux on macOS or Linux, native Windows is unsupported, and launching a rig writes provider hooks, workspace trust settings and configuration into ~/.claude.json and ~/.codex/config.toml — with an explicit warning to back up files first, since some writers recover unreadable settings as empty objects and no complete rollback is guaranteed. Version 0.6.0 dropped Node 20 because of its SQLite binding. Full-bypass permission modes exist but are off by default. The repository gained roughly 622 stars in a day.
