A terminal-based, open-source coding agent that connects to arbitrary LLM providers and executes code with OS-level sandboxing. It runs entirely on the local machine and provides an interactive TUI for managing codebase modifications.
Project overview
A terminal-based, open-source coding agent that connects to arbitrary LLM providers and executes code with OS-level sandboxing. It runs entirely on the local machine and provides an interactive TUI for managing codebase modifications.
Project type
AI Agent · AI Coding
Use cases
Coding & Development · Automation
Deployment
Refer to project documentation
License
MIT
Best for
Developers seeking a terminal-based coding agent that supports arbitrary LLM providers and executes tasks with local OS-level sandboxing.
Users who prefer a self-hosted, local-runnable tool that offers durable execution and side-git snapshots for managing code changes.
Key capabilities
Provides an interactive Terminal User Interface (TUI) with a context inspector, plan/act/operate cycles, and approval postures for interactive work.
Limitations and risks
Running entirely on your machine implies local hardware and resource dependencies for both the agent and the connected models.
Getting started
Install globally using npm, set your authentication provider, and launch the application to begin interacting with the agent.
Evidence and sources
README: A coding agent for your terminal. Works with any model; open models first.
README: npm install -g codewhale
README: codewhale auth set --provider deepseek # or export ANTHROPIC_API_KEY, etc. codewhale # open the TUI codewhale exec "fix the failing test" # headless
README: Talks to hosted open-model providers (`deepseek`, `openrouter`, `moonshot`, `zai`, `minimax`, `nvidia-nim`, …), to your own `vllm` / `sglang` / `ollama` with no key, and to Anthro…
README: Runs multiple workers durably: Fleet records work in an append-only ledger, so runs survive restarts and `fleet resume` picks up where things stopped.