run-kit is an open-source tmux orchestrator and remote management console developed by Sahil Ahuja as part of the shll.ai toolkit. It operates as an agent-agnostic harness, exposing active tmux sessions and panes through a browser-based dashboard accessible from a local workstation or mobile devices over networks like Tailscale. In daily use, run-kit provides terminal multiplexing workflows specifically suited for parallel development. The `run-kit riff` command provisions isolated agent workspaces across distinct Git worktrees (integrating with sibling tool `wt`), while `run-kit serve` launches a live web interface to view session outputs, host metrics, and pane states. Because it connects directly to tmux without wrapping application processes, run-kit is completely model- and agent-agnostic—equally supporting Claude Code, Aider, custom shell scripts, REPLs, and local development servers. What distinguishes run-kit from agent harnesses that wrap LLM APIs is its pure terminal multiplexer foundation. Rather than binding to a single agent protocol or proprietary execution loop, run-kit treats all AI coding agents as standard shell processes. This design ensures that agent orchestration remains durable even as underlying AI tools and CLI agents evolve. run-kit is free, open source under the MIT license, and distributed as a component of the broader shll terminal ecosystem.
Tags: agent harness, ai agent, cli, developer tools, terminal