Open Interpreter: A Coding Agent for Low-Cost AI Models

Open Interpreter is a coding agent optimized for low-cost models, forked from OpenAI's Codex to support various model providers. It can be quickly started on macOS, Linux, and Windows via an installation script, and sessions are run from the terminal using the i or interpreter command.

The key update is Kimi K3 support, ensuring top performance through a Rust-based implementation. The /harness command allows switching between multiple agent environments including native, claude-code, kimi-code, qwen-code, deepseek-tui, and swe-agent.

Key features:

• Command execution with native sandboxing on all OSes
• Switch providers and models in the TUI with /model
• Rust-based model harness inspection and switching
• Agent Client Protocol (ACP) compatibility for editor integration (interpreter acp)
• Computer use capabilities via agent-browser for web apps and trycua for native apps
• Local management of settings and session state in ~/.openinterpreter

Compatibility: If you're using OpenAI's Codex SDK, you can switch to Open Interpreter with just a one-line change. It offers full Codex exec protocol compatibility, which can be tested locally with scripts/test-codex-sdk-compat.sh.

Notably, this is a new Rust-based version, while the original Python project is maintained as a community fork at endolith/open-interpreter.