openinterpreter/openinterpreter

A coding agent for open models like Kimi K3

What it solves

Open Interpreter allows users to run a coding agent directly in their terminal, enabling LLMs to execute code and interact with the local system. It specifically focuses on maximizing the performance of low-cost models by emulating various agent harnesses.

How it works

It acts as a fork of OpenAI's Codex, implementing a protocol for executing code. It provides a set of interchangeable "harnesses" (such as Kimi Code, Claude Code, and DeepSeek) that optimize how the model interacts with the environment. It supports native sandboxing across macOS, Linux, and Windows and integrates with the Agent Client Protocol (ACP) and Model Context Protocol (MCP) for portability and editor integration.

Who it’s for

Developers and power users who want a local coding agent that is model-agnostic, portable, and capable of operating native applications or web apps via built-in QA skills.

Highlights

  • Multi-model support: Switch providers and models directly from the TUI using /model.
  • Harness Emulation: Use /harness to switch between different optimized agent harnesses to get the best performance from specific models.
  • Computer Use: Ability to drive web apps via agent-browser and native apps via trycua.
  • Portable Standards: Uses shared AGENTS.md instructions and .agents/skills directories to avoid vendor lock-in.
  • Native Sandboxing: Runs commands securely within native sandboxes on all major operating systems.

Related

  • Project
    0xku/kon

    A minimal coding agent harness with a tiny core prompt and a small toolset, designed for efficiency and transparency in terminal-based AI coding.

  • Project
    open-webui/open-terminal

    A self-hosted terminal server that provides AI agents and automation tools a sandboxed or bare-metal environment to execute code and manage files via a REST API.

  • Project
    Nano-Collective/nanocoder

    An open-source, local-first coding agent for the terminal that allows developers to use their own AI models via Ollama or various APIs while keeping code private.

  • Project
    sandbaseai/sandbase-harness

    A local-first runtime for AI agents providing sandboxed tool execution, persistent sessions, and audit trails for production-grade agent infrastructure.

  • Project
    ConardLi/easy-agent

    Easy Agent is a terminal‑based, TypeScript‑written coding assistant that streams chat with Claude, OpenAI, Gemini, Ollama and other LLMs, lets the model read/write files, run shell commands, browse the web, and be extended via plugins and multi‑agent teams. It is at release‑candidate stage, awaiting npm publication.