0xwilliamortiz/openclaude-improved

cd openclaude-main && bun install
bun run build
npm install -g .
openclaude

Usage

• •Run /provider inside OpenClaude for guided interactive setup (credentials saved to .openclaude-profile.json)
• •Or env vars: CLAUDE_CODE_USE_OPENAI=1, OPENAI_API_KEY=..., OPENAI_MODEL=gpt-4o, then openclaude
• •openclaude --continue / --resume <session-id> / --resume <session-id> --fork-session
• •openclaude --bg "fix failing tests"; openclaude ps / logs <name> / kill <name> for background sessions

Key features

• ✓20+ providers: OpenAI-compatible endpoints (OpenRouter, DeepSeek, Groq, Mistral, LM Studio), Ollama (local, no key), Gemini, GitHub Models (/onboard-github), Codex OAuth, Bedrock/Vertex/Foundry, Z.AI GLM, Fireworks, Cloudflare Workers AI, and more
• ✓Full toolset: Bash, read/write/edit, grep, glob, agents, tasks, MCP, slash commands; streaming, vision (where provider supports), repo map (PageRank, /repomap), per-agent provider/model routing
• ✓Bundled VS Code extension (launch integration, Control Center, in-editor chat); headless gRPC bidirectional server for CI and custom UIs
• ✓Background sessions as plain local child processes (no daemon); WebSearch via DuckDuckGo free by default with optional FIRECRAWL_API_KEY
• ✓Ollama integration talks native chat API with 32768-token window per request so same-session history isn't trimmed

Best for

Developers who want a Claude-Code-style CLI agent but prefer non-Anthropic models (OpenAI, local Ollama, OpenRouter, GitHub Models, etc.) with the same workflow.

Caveats

• ⚠Not affiliated with, endorsed by, or sponsored by Anthropic; 'Claude'/'Claude Code' are Anthropic trademarks; derived Claude Code code remains Anthropic's (MIT for modifications)
• ⚠Requires Node >=22 (enforced by engines.node) and Bun; built from source, no prebuilt binaries
• ⚠Does NOT read ~/.claude, project .claude/ dirs, or CLAUDE_CONFIG_DIR — owns ~/.openclaude/ and ~/.openclaude.json; migrate by copying only files you wrote
• ⚠Anthropic-only features don't exist on every backend; tool quality tracks model quality (small local models struggle with long multi-step tool loops); some providers cap output below CLI defaults
• ⚠Project .env files are not auto-loaded (use --provider-env-file .env); ripgrep must be installed system-wide