What is Open Interpreter?

Open Interpreter is a terminal coding agent optimized for low-cost and open models. It is a fork of OpenAI's Codex that can emulate other agent harnesses (such as Claude Code, Kimi Code or Qwen Code) to get the best results from each model, runs commands in native sandboxes on macOS, Linux and Windows, and works with ACP-compatible editors and the Codex SDK.

Is Open Interpreter open source?

Yes, licensed under Apache-2.0.

Open Interpreter

Active·★ 68.5k·Apache-2.0·Updated 2026-10-07

★ Most Popular★ Trending★ Essential

Open Interpreter is a terminal coding agent built to get the best results out of low-cost and open models. The project has been rebuilt as a Rust fork of OpenAI's Codex, and its distinctive feature is harness emulation: it can mimic the agent loop of tools such as Claude Code, Kimi Code, Qwen Code or SWE-agent, so each model runs inside the harness it performs best in.

It suits developers who want a capable coding agent without being locked into one premium model, for example teams running Kimi, GLM, Qwen or DeepSeek models, or anyone who switches providers often. It runs commands inside native sandboxes on macOS, Linux and Windows, can be used from ACP-compatible editors, acts as a drop-in replacement for the Codex SDK, and reads shared AGENTS.md files and .agents/skills directories so your setup stays portable. A built-in QA skill can drive web apps in a real browser or test native apps.

Getting started: install with `curl -fsSL https://www.openinterpreter.com/install | sh` on macOS and Linux, or the PowerShell script on Windows, then type `i` or `interpreter` to start a session. Use `/model` to choose a provider and `/harness` to switch harnesses.

#LLM#Code Execution#AI Agent#Python#Local AI#Coding

Features

01Switch between emulated agent harnesses (native, Claude Code, Kimi Code, Qwen Code, SWE-agent and more) with /harness

02Native command sandboxing on macOS, Linux and Windows

03Switch providers and models from the TUI, including any OpenAI-compatible endpoint

04Runs as an Agent Client Protocol (ACP) agent for editors and as a drop-in for the Codex SDK

05Supports MCP, skills, hooks, permissions and shared AGENTS.md instructions

Why choose it

+Harness emulation lets cheaper and open models perform closer to their best, instead of forcing every model through one generic loop.

+Provider-agnostic: switch models from the TUI and use any OpenAI-compatible endpoint.

+Portable by design: uses shared standards (AGENTS.md, .agents/skills, MCP, ACP, Codex exec protocol) rather than a proprietary format.

+Native sandboxing on all three desktop platforms, and Apache-2.0 licensed.

Trade-offs

!The current project is a complete change from the original Python "let LLMs run code locally" tool; older tutorials, the old pip package and Python integrations no longer describe how it works.

!Results depend heavily on the model and harness you pick, so expect some experimentation to find a good combination.

!Fast-moving project built around the latest model releases; behaviour and defaults can change quickly.

Compatibility

macOS

Native

Verified via docs

Linux

Supported

Verified via docs

Windows

Partial

Verified via docs

Python 3.10+

Verified via docs

Quick start

$ curl -fsSL https://www.openinterpreter.com/install | sh

Use cases

↳Getting strong coding-agent results from inexpensive or open models such as Kimi, GLM, Qwen or DeepSeek

↳Using one terminal agent across several model providers without changing tools

↳Testing web and native apps with the built-in QA skill

How it compares

≈vs OpenAI Codex CLI: Open Interpreter is a fork of Codex and keeps its exec protocol, but focuses on running non-OpenAI, low-cost and open models well.

≈vs Gemini CLI: Gemini CLI is the first-party agent for Gemini models; Open Interpreter is model-agnostic and can switch harness per model.

≈vs Claude Code: rather than competing head-on, Open Interpreter can emulate the Claude Code harness for other models.

Alternatives

Context7★ 62.8k

MCP Server that provides up-to-date code documentation for LLMs and AI code editors.

vs →

Gemini CLI★ 107.3k

An open-source AI agent that brings the power of Gemini directly into your terminal. Supports native MCP.

vs →

ragflow★ 91.9k

RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context layer for LLMs

GitHub MCP Server★ 33.5k

GitHub's official MCP Server. Allows AI agents to interact directly with your GitHub repositories (read files, search code, issues).

n8n★ 206.7k

Fair-code workflow automation platform with native AI capabilities. Combine visual building with custom code, self-host or cloud, 400+ integrations.

Brave Search MCP★ 1.5k

Allow your AI Agent to search the real-time internet using Brave Search API. Essential for getting up-to-date information.

Microsoft AutoGen★ 61.3k

A framework that enables the development of LLM applications using multiple agents that can converse with each other to solve tasks.

CrewAI★ 59.5k

Framework for orchestrating role-playing, autonomous AI agents. By working together, your Crew can tackle complex tasks.

See all alternatives →

Comments

No comments yet. Be the first!

Open Interpreter

Active·★ 68.5k·Apache-2.0·Updated 2026-10-07

★ Most Popular★ Trending★ Essential

#LLM#Code Execution#AI Agent#Python#Local AI#Coding

Features

01Switch between emulated agent harnesses (native, Claude Code, Kimi Code, Qwen Code, SWE-agent and more) with /harness

02Native command sandboxing on macOS, Linux and Windows

03Switch providers and models from the TUI, including any OpenAI-compatible endpoint

04Runs as an Agent Client Protocol (ACP) agent for editors and as a drop-in for the Codex SDK

05Supports MCP, skills, hooks, permissions and shared AGENTS.md instructions

Why choose it

+Harness emulation lets cheaper and open models perform closer to their best, instead of forcing every model through one generic loop.

+Provider-agnostic: switch models from the TUI and use any OpenAI-compatible endpoint.

+Portable by design: uses shared standards (AGENTS.md, .agents/skills, MCP, ACP, Codex exec protocol) rather than a proprietary format.

+Native sandboxing on all three desktop platforms, and Apache-2.0 licensed.

Trade-offs

!The current project is a complete change from the original Python "let LLMs run code locally" tool; older tutorials, the old pip package and Python integrations no longer describe how it works.

!Results depend heavily on the model and harness you pick, so expect some experimentation to find a good combination.

!Fast-moving project built around the latest model releases; behaviour and defaults can change quickly.

Compatibility

macOS

Native

Verified via docs

Linux

Supported

Verified via docs

Windows

Partial

Verified via docs

Python 3.10+

Verified via docs

Quick start

$ curl -fsSL https://www.openinterpreter.com/install | sh

Use cases

↳Getting strong coding-agent results from inexpensive or open models such as Kimi, GLM, Qwen or DeepSeek

↳Using one terminal agent across several model providers without changing tools

↳Testing web and native apps with the built-in QA skill

How it compares

≈vs OpenAI Codex CLI: Open Interpreter is a fork of Codex and keeps its exec protocol, but focuses on running non-OpenAI, low-cost and open models well.

≈vs Gemini CLI: Gemini CLI is the first-party agent for Gemini models; Open Interpreter is model-agnostic and can switch harness per model.

≈vs Claude Code: rather than competing head-on, Open Interpreter can emulate the Claude Code harness for other models.