Switch between cloud and local AI models in a single terminal for coding tasks
Imagine typing a command in your terminal and watching a chosen AI model read, modify, and search the web for code. OpenClaude is an open-source CLI tool that unifies various AI services like OpenAI, Gemini, and Ollama, along with local models, into a single interface. It avoids dependency on a specific cloud provider, allowing you to select from models installed on your computer or multiple APIs as needed.
One terminal, various AI engines
OpenClaude supports various backends, including OpenAI-compatible APIs, Gemini, GitHub Models, Codex, and Ollama. Running the /provider command lets you set up a provider and save a profile following the prompts, eliminating the need to manually configure complex environment variables. Since it supports both cloud APIs and local model servers, you can switch between local inference that requires no internet connection and the latest cloud models within a single workflow. Specifically, when using Ollama, it automatically adjusts the context window to prevent conversation history from being truncated.
Functionality as a coding agent
This tool is an agent that performs actual coding tasks rather than just chatting. It handles multi-step tasks by invoking various tools such as Bash execution, file reading/writing/editing, grep, glob, web search, and MCP (Model Context Protocol). You can monitor token generation and tool execution progress in real time through streaming output. Additionally, the 'Repo Map' feature, which maps repository structure by PageRank importance, automatically injects the core structure of the codebase into the context. A VS Code extension is also included, providing editor integration and theme support.
Session management and background execution
For long-running tasks, you can execute them in a background session detached from the terminal using the --bg option. You can check running sessions with openclaude ps, track logs with openclaude logs, and terminate sessions with openclaude kill. You can also restore existing conversations by session ID or fork the conversation history into a new session using --fork-session. These features operate as local child processes and do not start separate daemons or network services.
Installation and requirements
Installing OpenClaude requires Node.js >=22.0.0. You can install it via npm using the command npm install -g @gitlawb/openclaude@latest, and Arch Linux users can also install it via the AUR package. After installation, if ripgrep is missing, it must be installed on the system. Building from source or local development also requires Bun 1.3.13 or higher. Default settings are stored in the ~/.openclaude directory, and it does not automatically read existing Claude Code settings (~/.claude), allowing for independent use.
Pre-use considerations
OpenClaude is an independent community project unrelated to Anthropic, derived from and modified based on the Claude Code codebase. The license is MIT, but the derived Claude Code portions are owned by Anthropic. It does not guarantee identical functionality across all providers, and specifically, Anthropic-exclusive features may not work with other providers. Small local models may struggle with long multi-step tool flows, so it is recommended to use models with strong tool-calling capabilities. Additionally, web search uses Exa's free tier by default, but due to limitations, you can set an API key or choose a different backend.