openinterpreter

A lightweight coding agent for open models like Deepseek, Kimi, and Qwen

⭐ 64,091 stars🍴 5,555 forksRustUpdated 6/20/2026localterminalcoding-agentdeepseekinterpreterkimi

Start hereRead the setup guide →Install steps and usage, straight from the docs.Source codeDownload ZIPLatest commit on default branch.SourceView on GitHubgithub.com/openinterpreter/openinterpreter

Project site: openinterpreter.com/

About openinterpreter

openinterpreter is a Rust open-source code generation project by openinterpreter. A lightweight coding agent for open models like Deepseek, Kimi, and Qwen With 64,091 GitHub stars and 5,555forks, it's one of the Code Generation tools worth knowing. You can download the source, browse it on GitHub, or read the full setup guide in the README below.

A coding agent that runs where your files are

Open Interpreter gives a language model a terminal on your own computer. You describe what you want in plain language; it writes code, runs it locally, reads the result, and iterates until the task is done. Because it executes on your machine rather than in a hosted sandbox, it can touch the things a sandbox cannot — your actual files, your installed tools, your local network.

Built for open models

The project is aimed squarely at open-weight models such as Deepseek, Kimi and Qwen rather than assuming a single commercial provider. That matters for two reasons: cost, since a long agent session makes many calls, and privacy, since a model you run yourself never sees your codebase leave the building. It is deliberately lightweight — a terminal tool, not an IDE.

Read this before you run it

The capability and the risk are the same thing. An agent that can execute arbitrary code on your machine can also delete files, exfiltrate data, or act on instructions hidden in a file it reads. Run it against a directory you have backed up, prefer a container or a virtual machine for anything unfamiliar, and keep approval prompts on until you trust a given task. Treat "it wrote the code itself" as a reason for more scrutiny, not less.

When to pick something else

If you want a coding assistant inside your editor with no execution, an IDE extension is a better fit. Open Interpreter earns its place when the job is genuinely agentic — running commands, inspecting output, and deciding what to do next.

From the project README

Open Interpreter

A coding agent optimized for low-cost models. Blog post ↗

English • Español • 简体中文

[!NOTE]

Today: Kimi K3 is here. We have reimplemented the provider-recommended

Kimi Code harness in Rust, giving you

maximum K3 performance with a Codex-like interface.

Kimi Docs →

<br>

This is an excerpt from the project's own README, quoted here for reference. Read the full README on GitHub →

openinterpreter — FAQ

Is Open Interpreter safe to run?

It executes code on your own machine, so treat it with the caution that implies. Run it in a container or virtual machine for unfamiliar tasks, point it only at directories you have backed up, and keep confirmation prompts enabled until you trust the workflow.

Which models does Open Interpreter work with?

It is designed around open-weight models such as Deepseek, Kimi and Qwen, which you can run locally or call through a provider. That focus keeps long agent sessions affordable and lets your code stay on your own hardware.

How is it different from a coding assistant in my editor?

An editor assistant suggests code for you to accept. Open Interpreter runs it — executing commands, reading the output and deciding the next step. That makes it more capable for multi-step tasks and correspondingly more consequential when it gets something wrong.

Does my code get sent to a server?

That depends entirely on the model you point it at. Running a local open-weight model keeps everything on your machine; pointing it at a hosted API sends whatever context the agent gathers to that provider.