What it does

Claw Code provides a claw CLI binary that wraps LLM APIs (Anthropic, OpenAI, local Ollama/llama.cpp/vLLM) into an interactive coding agent harness. You clone the repo, build the Rust workspace, set an API key, and run claw prompt or claw doctor to validate your setup. It handles terminal navigation, file context via @path references, session management, and a parity harness for testing against reference behavior.

The interesting bit

The project is deliberately not the official Claude Code — it is a clean-room Rust port with its own crate structure, a mock parity harness for deterministic testing, and a container-first workflow. The README is refreshingly honest about what is missing: no ACP/Zed daemon yet, no cargo install convenience (the crates.io name is a deprecated stub), and a whole PARITY.md file tracking how far the Rust version has caught up.

Key highlights

• Build-from-source only; canonical code lives in rust/ workspace
• Multi-provider support: Anthropic, OpenAI, and local OpenAI-compatible servers
• claw doctor health-checks your API key, model access, and tool config
• Windows-first PowerShell documentation alongside Unix paths
• Deterministic mock-service parity harness for testing without live credentials
• MIT licensed, with explicit disclaimed affiliation to Anthropic

Caveats

• cargo install claw-code installs a deprecated stub that prints a rename notice; you must build from source or cargo install agent-code for the upstream binary
• ACP/Zed editor integration is not yet implemented; claw acp serve is currently just a status alias
• The src/ and tests/ directories contain companion Python/reference code, not the primary runtime

Verdict

Worth a look if you want a hackable, local-first alternative to the official Claude Code binary and are comfortable compiling Rust. Skip it if you need a polished one-click install or mature editor daemon support.