Build and run an AI coding agent locally in your terminal to automate coding tasks.
Study the open-source Rust implementation of an agent harness and contribute improvements.
Integrate Claude API access into your development workflow via command-line interface.
Requires Anthropic API key and Rust/Cargo installation.
Claw Code is, according to its README, "the public Rust implementation of the claw CLI agent harness." A CLI agent harness here is a command-line tool that talks to an AI model and lets the model take actions on your machine, the README's quick start shows commands like claw prompt "say hello" and a claw doctor health check. The canonical source lives in the rust/ workspace inside this repo, and the project description calls it "the fastest repo in history to surpass 100K stars." The way it works is that you clone the repo, run cargo build --workspace inside rust/, and then run the resulting claw binary, having first set an API key in the environment. The README explicitly says claw requires an API key (it lists ANTHROPIC_API_KEY and OPENAI_API_KEY as examples), a Claude subscription login is not a supported auth path. After building, the binary lives at rust/target/debug/claw (or claw.exe on Windows), there are instructions for symlinking it onto PATH, using cargo install --path ., or editing your shell profile. A separate test suite runs via cargo test --workspace. The README also warns that cargo install claw-code from crates.io is a deprecated stub and should not be used, the upstream binary is installed via cargo install agent-code instead. Someone would use this when they want to run a local AI coding agent from the command line and prefer the Rust build of the harness, or when they want to read and modify the harness source themselves. The repo also contains companion documentation files (USAGE.md, PARITY.md, ROADMAP.md, PHILOSOPHY.md, container.md) and a Python reference workspace.
This repo across BitVibe Labs
Verify against the repo before relying on details.