Claw Code

Claw Code is a Rust-based command-line AI agent harness for running interactive, tool-enabled coding and automation workflows from your terminal. Configure your preferred API provider, start a session, and let Claw help you explore projects, work with files, and execute agent-driven tasks.

Built around the claw CLI, Claw Code provides a practical local workflow for developers who want direct control over how they run AI agents. Use a prompt for one-off requests, open an interactive session for longer work, and run built-in checks to confirm that credentials, model access, and tool configuration are ready.

Features Terminal-first AI agent workflow — Run prompts and interactive agent sessions directly from the command line

Rust implementation — A canonical Rust workspace designed for fast, native command-line use

Multi-provider flexibility — Supports API-key-based model configuration, including Anthropic, OpenAI-compatible services, and documented local-provider workflows

Project-aware assistance — Navigate files, provide file context, and work within your local development environment

Health checks — Use claw doctor to validate configuration, credentials, model availability, and tools

Cross-platform setup — Build and run on macOS, Linux, and Windows, including a PowerShell-first Windows path

Container-friendly — Documentation and configuration paths support container-oriented workflows

Open source — Distributed under the MIT License

Get started by building the Rust workspace, setting an API key, and running your first prompt.

Claw Code is best suited to developers, tinkerers, and AI-tooling enthusiasts who prefer a transparent, command-line-centric workflow over a closed desktop app. It is a build-from-source project, so Rust and a compatible model-provider API key are required.

Important: Claw Code describes itself as an experimental, agent-managed exhibit rather than a conventional production product. For production-oriented agent workflows, the project directs users toward its related Gajae-Code and LazyCodex harnesses.