Build a sustainable development flow by uniting Claude Code and Codex into a single team
Imagine moving away from juggling multiple terminal windows to run agents, toward a scene where you form and manage a team with a single command. OpenRig is a multi-agent harness that binds individual agents like Claude Code and Codex into a single system. Tell the lead agent the result you want, and specialized agents collaborate to bring it back, requesting decisions as needed.
Defining agent team structure with YAML
OpenRig declaratively defines the composition of an agent team using a YAML file (RigSpec). This file includes the groups (Pods) agents belong to, the connections between agents (Edges), and policies (Continuity policies) that maintain state even after a session reboot. A single rig up command automatically handles everything from the tmux session and harness to the startup file and readiness check. You can also find existing Claude Code or Codex sessions running and include them in management using rig discover.
Terminal UI and inter-agent communication
You can monitor the team's topology, projects, feed, and system health in real time through the TUI (Terminal User Interface). Agents collaborate by exchanging messages using the rig send, rig broadcast, and rig chatroom commands. If you need to change the structure of a running team, you can add or remove seats using rig grow or rig shrink. Since all agents run in tmux sessions, you can attach directly to inspect or intervene in work when necessary.
System impact and security settings
OpenRig applies several changes to system files during installation and execution. It adds mouse support settings to ~/.tmux.conf and creates instance state and a database in the ~/.openrig directory. It also records trust settings and activity hooks in the configuration files for Claude Code and Codex. These changes are intended to manage the permissions and context required for agents to perform tasks. Before the first run, you can preview changes with rig setup --dry-run, and backing up related files is recommended. Specifically, Codex sets the trust level for the workspace to 'trusted', while Claude Code applies edit-permission mode by default.
Execution environment and requirements
To run OpenRig, you need Node.js 20, 22, or 24, and tmux. The operating systems supported are macOS and Linux, with Node.js 22 recommended for Apple silicon Macs. Native Windows is not yet supported, and WSL2 has not been tested. Running a service-based Rig additionally requires Docker. You can optionally use herdr or cmux to view the terminal workspace together. For installation, use npm install -g @openrig/cli or install via Bun.
Caution regarding system file changes
OpenRig automates system settings to allow agents to work safely, but this directly affects the user's local file system. In particular, it writes trust records and hook settings to the configuration files for Claude Code and Codex, which may overwrite or modify existing settings. The README states that these changes are automated and that no interactive preview is provided. Therefore, you must back up the relevant configuration files before first use. Additionally, previously running sessions may need to be restarted to reflect the newly written runtime settings. You should use these features only after fully understanding these changes to ensure system stability and data integrity.