What it does
Open Interpreter is a coding agent that runs in your terminal and is aimed at low-cost and open models such as Kimi K3 and GLM 5.3. The repository calls it the new Rust version of Open Interpreter, a fork of OpenAI's Codex, and notes that the original Python project lives on as a community-maintained fork. Its main idea is harness emulation. The /harness command switches the agent between styles modelled on other coding agents, with options including claude-code, kimi-code, qwen-code and swe-agent, to get the best results from cheaper models. The /model command switches providers and models, and any OpenAI-compatible provider can be used through Chat Completions. Commands run inside native sandboxing on macOS, Linux and Windows. The agent supports MCP, skills, hooks, permissions and shared AGENTS.md instructions, and it can act as an Agent Client Protocol agent in compatible editors with interpreter acp. A built-in QA skill lets a model drive web apps in a real browser or test native apps. Because it speaks the Codex exec protocol, existing Codex SDK code can use it with a one-line binary override. A one-line shell installer is given for macOS and Linux and a PowerShell command for Windows. You then start a session by typing i or interpreter. The project is licensed under Apache-2.0.
Capabilities
- Switch agent harnesses with /harness, including claude-code, kimi-code and qwen-code styles
- Switch providers and models from the terminal with /model
- Native sandboxing for commands on macOS, Linux and Windows
- Runs inside Agent Client Protocol editors with interpreter acp
- Built-in QA skill for testing web and native apps
- Supports MCP, skills, hooks, permissions and AGENTS.md
When to use it
- You want a terminal coding agent that works with lower-cost open models
- You want to compare how different agent harnesses behave on the same model
- You already use the Codex SDK and want to swap in a different binary
- You want a coding agent inside an editor that supports the Agent Client Protocol
