OpenRig is a self‑hosted system that turns individual AI coding agents (Claude Code and Codex) into a coordinated, persistent team managed through a declarative YAML rig specification. After installing the CLI globally with Node 20, 22 or 24 (via npm or Bun) and ensuring tmux is present, users run rig setup to probe required tools, optionally install cmux, and write an OpenRig block to ~/.tmux.conf for mouse support and scrollback. The daemon creates an instance state directory at ~/.openrig , seeding a SQLite database, plugin resources, and the openrig-skills discovery skill in Claude and Codex configuration folders. During setup it also writes Codex hook configuration and trust records, and configures Claude’s workspace trust ( ~/.claude.json , .claude/settings.local.json ) and MCP resources. Launching a rig with rig up <name> spawns tmux sessions for each seat, injects identity and connection environment variables ( HOME , CODEX_HOME , OPENRIG_* ), and pre‑trusts the workspace. Agents communicate via rig send , rig broadcast , or rig chatroom , while the TUI ( rig tui --shared ) displays topology, seat details, and system health. Snapshots ( rig down --snapshot ) capture full state for later restoration ( rig up <name> ), and operators can scale teams with rig grow , rig shrink , rig launch , or rig remove . Optional integrations include herdr or cmux for shared terminal views, Docker for service‑backed rigs, and a MCP server that lets agents manipulate the rig’s topology directly. All operations are logged, and the system preserves state across reboots, preventing terminal sprawl while maintaining explicit permission controls and trust management.