agents-libagents-lib
AgentsMCP ServersWorkflowsTutorialsComparisonsTemplatesNewsAbout
OP

OpenInterpreter

⭐ 65,790•Rust•coding-agent

Coding agent optimized for low-cost models, rebuilt in Rust with native sandboxing.

GitHubWebsite →
RustCoding AgentLow-Cost ModelsSandboxingMCPComputer Use

Overview

OpenInterpreter is a coding agent optimized for low-cost models with over 65,000 GitHub stars. The Rust-native rewrite brings native sandboxing across macOS, Linux, and Windows, multi-harness support for different model families, MCP integration, computer use via browser automation, and the Agent Client Protocol for IDE integration.

Features

  • ✓Multi-harness support (native, Claude Code, ZCode, SWE-agent, more)
  • ✓Native sandboxing across macOS, Linux, and Windows
  • ✓MCP server integration
  • ✓Computer use via browser automation
  • ✓Agent Client Protocol (ACP) for IDE integration
  • ✓Skills and hooks ecosystem

Installation

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

Pros

  • +Built in Rust — fast, memory-safe, cross-platform
  • +Works with low-cost models effectively
  • +Native sandboxing across all major operating systems
  • +Multiple harness modes for different model families
  • +Full MCP and skills ecosystem compatibility

Cons

  • −New Rust version with some features still maturing
  • −Primarily terminal-focused — not GUI-heavy
  • −Documentation still catching up with feature velocity

Alternatives

Claude Code →OpenHands →SWE-agent →

Documentation

OpenInterpreter

Overview

OpenInterpreter is a coding agent optimized for low-cost models, rebuilt from the ground up in Rust. With over 65,000 GitHub stars, it has evolved from a Python-based prototype into a production-grade terminal agent with native sandboxing across macOS, Linux, and Windows. Originally inspired by OpenAI Codex, OpenInterpreter now supports multiple agent harnesses and model providers, making it one of the most flexible coding assistant runtimes available.

The Rust-native rewrite dramatically improves performance, memory safety, and cross-platform reliability. It runs directly in your terminal and can execute shell commands, edit files, run tests, and interact with browsers — all within a sandboxed environment.

Features

  • Multi-Harness Support: Native, Claude Code, Claude Code Bare, ZCode, Kimi CLI, Qwen Code, DeepSeek TUI, SWE-agent, and minimal harness modes
  • Native Sandboxing: Commands execute in isolated environments on macOS, Linux, and Windows
  • MCP Integration: Full support for Model Context Protocol servers and tools
  • Computer Use: Browser automation capabilities for end-to-end web interactions
  • Agent Client Protocol (ACP): Editor integration for IDE workflows
  • Skills & Hooks: Extensible skill system and lifecycle hooks (PreToolUse, PostToolUse)
  • Permissions System: Granular access control for tool and command execution
  • Model Flexibility: Switch providers and models via /model command — works with any OpenAI-compatible API
  • AGENTS.md Support: Native integration with project-level agent instructions

Installation

macOS / Linux:

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

Windows (PowerShell):

irm https://www.openinterpreter.com/install.ps1 | iex

Launch with i or interpreter in your terminal.

Quick Start

# Launch the interactive agent
i

# Switch to a different model
/model claude-opus-4-7

# Switch harness mode
/harness claude-code

# Run a command
Run tests for the auth module and fix any failures

Core Concepts

Agent Harness

A harness defines how the agent loops, calls tools, and manages state. Different harnesses are optimized for different model families and use cases. The native harness is designed for cost-efficient models, while Claude Code harness provides compatibility with Anthropic's agent patterns.

Sandboxing

Commands execute within OS-native sandboxing mechanisms — containers on Linux/macOS and job objects on Windows. This isolates file system access and network calls while maintaining a natural terminal experience.

Skills

Skills are markdown files that define reusable workflows. OpenInterpreter is compatible with the anthropics/skills standard, making skills portable across Claude Code and OpenInterpreter.

Advanced Features

Custom Harness Configuration

# Configure a custom model provider
/model https://api.openrouter.ai/v1 deepseek-r1

# Set custom tool permissions
/permissions allow network deny filesystem

MCP Server Integration

# Register an MCP server
/mcp add filesystem --command npx --args @modelcontextprotocol/server-filesystem

Computer Use (Browser Automation)

# Navigate and interact with websites
/mode computer
Navigate to github.com and find trending repositories

Examples

  • Debug a Failing Build: "Find and fix the compilation errors in the src directory"
  • Refactor Codebase: "Rename all occurrences of userService to userClient across the project"
  • Run Tests & Fix: "Run the full test suite and fix any failing tests with explanations"
  • Explore Code: "Find all API endpoints that handle user authentication"

Pros

  • ✅ Built in Rust — fast, memory-safe, cross-platform
  • ✅ Works with low-cost models without sacrificing capability
  • ✅ Native sandboxing across all major operating systems
  • ✅ Multiple harness modes for different model families
  • ✅ Full MCP and skills ecosystem compatibility
  • ✅ Large and active community (65K+ stars)

Cons

  • ❌ New Rust version means some features are still maturing
  • ❌ Primarily terminal-focused — not ideal for GUI-heavy workflows
  • ❌ Less polished than commercial IDE-integrated assistants
  • ❌ Documentation still catching up with feature velocity

When to Use

Use OpenInterpreter when you need a cost-effective, sandboxed coding agent that works across multiple model providers. Ideal for developers who want CLI-first agent workflows, need to run agents on local or private models, or want maximum flexibility in agent harness selection.

Resources