# Open Interpreter

Open Interpreter is a coding agent forked from OpenAI's Codex and tuned to get strong results from lower-cost open models such as Kimi K3 and GLM.

It switches between several "harness" emulations, including a Claude-Code-like mode, a Kimi-specific mode and others, to match whichever model is doing the work, and speaks the same exec protocol as the Codex SDK, so an existing Codex-based integration can point at it with a one-line change. It is also compatible with editors that use the Agent Client Protocol.

Configuration favors shared, tool-neutral standards over a product-specific format: it reads repository AGENTS.md files and shared .agents/skills directories alongside MCP, keeping only runtime state and legacy skill directories under its own product-specific folder. A built-in QA skill lets any connected model drive and test web or native apps through a real browser or a desktop-automation tool.

Runs native sandboxing on macOS, Linux and Windows. Rust, Apache-2.0.

Website: http://openinterpreter.com/