Show HN: Ouijit, command terminals running coding agents
Ouijit is a free, open-source tool that runs coding agents inside terminal sessions, with built-in sandboxing via Lima VMs, session-scoped CLI awareness for agents, task lifecycle hooks, and support for organizing child terminals, markdown, and URL previews. It integrates with Claude Code, Codex, Pi, and OpenCode, requires no login or telemetry, and works on macOS and Linux.
Background
- Ouijit is a new open-source terminal app (like iTerm or tmux) designed for software engineers who want to run AI coding agents (Claude Code, Codex, etc.) directly in the command line, rather than through a web chat interface.
- Most existing "agent orchestrator" tools force you into a chat UI, require an account login, or only work with the company's own model. Ouijit aims to be local-first, free, and model-agnostic.
- Key technical features: it can sandbox each coding task in a separate Linux VM (using Lima) to prevent conflicts; agents running inside it automatically get a CLI toolkit to control the terminal; and tasks have lifecycle hooks (e.g., run a script when a task is marked "done").
- The name evokes a "ouija board" — a playful reference to communicating with AI agents as if they were spirits.