OpenRig

Local daemon and CLI that boots persistent teams of coding-agent seats in tmux from YAML, with a TUI, work queue and MCP server.

Core features

• Sandboxed runs Partial

  OpenRig 0.7.8 ships a reference driver for an OpenShell sandbox, but its changelog says the worker never became usable in release testing. Docker is only used for services, not agents.

  Source
• Diff reviewNo

  README, docs/reference and the TUI and web UI code were searched for diff, git diff and changes. Neither interface shows code diffs.

• Agent coordination YesSource
• Scheduled tasks Yes

  Taken from the CLI code. A watchdog sends a message to a running seat at a set interval, and queue timers wake a seat once. It has no cron syntax.

  Source
• Event triggers Partial

  Slack messages can land in a seat's queue, but docs and code show no webhook, issue or pull request trigger.

  Source
• Team collaboration Partial

  Several people can be registered to talk to one rig over Slack, but there are no user logins or enforced roles. Policy text is read by agents, not enforced by OpenRig.

  Source
• API / CLI control YesSource

Pricing

Apache-2.0, no paid tier or pricing page found.

Platforms

macOS, Linux, WSL2 on Windows; needs Node.js 22 or 24 and tmux; older React web UI is in maintenance mode

Research notes

Own subscription: README says "no second subscription is required" and uses claude auth login and codex login. BYOK evidence covers Pi seats only (openrouter, zai, kimi-coding, minimax keys in the daemon environment); key handling for Claude and Codex seats is left to those CLIs, though help.md mentions the Codex API-key entry prompt. Deployment is local: README calls it a local daemon. Docs mention a remote daemon address and SSH handoff, but one reported Windows/WSL run lists cross-host operation as not tested.

Agents & providers

Codex Claude Code Pi ACP agents via rig acp run community harness presets