OpenClaude

OpenClaude is a fork of the leaked Claude Code source that replaces Anthropic's proprietary LLM backend with an OpenAI-compatible provider shim. It lets developers use all of Claude Code's powerful agentic tools — bash, file read/write/edit, grep, glob, MCP, agents, tasks — powered by any model that speaks the OpenAI chat completions API. Supported providers include OpenAI (GPT-4o), DeepSeek, Google Gemini (via OpenRouter), Llama, Mistral, Azure OpenAI, Groq, Together AI, Ollama, LM Studio, and the ChatGPT Codex backend. The project is written in TypeScript and installable via npm or buildable from source using Bun.

• OpenAI-compatible shim — Set three environment variables (CLAUDE_CODE_USE_OPENAI=1, OPENAI_API_KEY, OPENAI_MODEL) and run openclaude to get started with any supported provider.
• Full tool support — All Claude Code tools work out of the box: Bash, FileRead, FileWrite, FileEdit, Glob, Grep, WebFetch, WebSearch, Agent, MCP, LSP, NotebookEdit, and Tasks.
• Streaming & multi-step tool chains — Real-time token streaming and multi-step tool calling are fully supported across all compatible models.
• Local model support — Run entirely offline with Ollama or LM Studio; no API key required for local providers.
• Codex backend support — codexplan and codexspark aliases map to ChatGPT Codex backend models; reads ~/.codex/auth.json automatically if present.
• Provider launch profiles — Use bun run profile:init to bootstrap a provider profile and bun run dev:profile to launch with persisted settings, avoiding repeated environment setup.
• Runtime hardening tools — Built-in doctor:runtime, smoke, and hardening:check commands validate provider configuration and environment before launch.
• Vision & image support — Base64 and URL images are passed through to vision-capable models.
• Sub-agents & memory — AgentTool spawns sub-agents using the same configured provider; a persistent memory system is included.
• Slash commands — /commit, /review, /compact, /diff, /doctor, and more are all available.