# OpenHarness

A terminal AI coding agent that works with any LLM -- free local models through Ollama or cloud APIs -- with slash commands, checkpoints, hooks, agent roles and a headless mode for CI.

OpenHarness starts by auto-detecting a local Ollama installation and chatting with no API key needed, or can connect to a cloud provider instead. Every AI edit is git-committed and reversible with an undo command, and in-session commands can rewind the last change, list agent specializations, toggle vim mode, or flush the transcript to scrollback for review.

Its terminal UI is a sequential renderer inspired by Ink and Claude Code's default mode: completed messages flush to native, scrollable output, while a live area rewrites in place for streaming text, spinners and input. Companion Python and TypeScript SDKs let other programs -- notebooks, batch scripts, VS Code extensions, Electron apps -- drive it programmatically, with streaming events, stateful sessions, custom tools, a permission callback and session resume.

Website: https://github.com/zhijiewong/openharness
