Connects to a local LLM and works autonomously. Safe, private and fast.
Ultra lightweight
~1k tokens for system prompt, chat template, and tool schemas combined. Almost instant decode.
Fully local
Works with any OpenAI-compatible endpoint — Llama.cpp, Ollama, LM Studio, vLLM. Nothing leaves your machine.
Minimal code
The codebase is small and understandable. You can see exactly how Bare agentically works.
Safe by default
Per-tool permission prompts, read-only mode, and folder sandboxing. You control what Bare is allowed to do.
Plugin tools
Drop a .js file into plugins/ and Bare uses it instantly. Ask Bare to make you any tool. Basic tools come built in.
Context management
Context is automatically truncated so Bare keeps working (Only in llama.cpp, other servers use their own methods).
Install
Download the latest release from GitHub. Windows, macOS, and Linux.
Start a local LLM server, launch Bare, and go.