OpenHarness coordinates coding agents across machines

OpenHarness is a new open-source workspace for running multiple coding-agent CLIs across local and remote machines. It combines a desktop app, a background daemon on each host, an end-to-end encrypted relay, persistent tmux sessions, domain-specific tool packages, and an optional desk display. The project addresses a practical limitation of terminal-based agents: long tasks remain tied to individual shells and machines, complicating parallel work and remote supervision.

Sessions that survive disconnects

OpenHarness assigns each agent a persistent tmux session, allowing its process to continue after the desktop app closes or a connection drops. If a reboot stops tmux itself, the daemon recreates the panes and resumes the saved agent sessions.

The daemon integrates with supported agents through their existing transcripts and vendor-provided hooks. The current roster includes Claude Code, Codex, Cursor, OpenCode, Pi, Hermes, Command Code, Devin, Muse Code, Amp, Kilo, Grok Build, Antigravity, and GitHub Copilot. Credentials remain in directories such as ~/.claude and ~/.codex, so each engine continues using its existing subscription or API key.

Remote control through outbound links

Each linked machine runs a daemon that initiates outbound connections. Terminal traffic is encrypted end to end, and the relay forwards ciphertext. When the network permits, the desktop app and daemon establish a direct peer-to-peer connection over WebRTC. The design needs no SSH server, VPN, or public inbound port.