hadihonarvar/flock

Flock is a self-hosted service written in Go that sits between your team's AI coding tools and the language models they talk to. Instead of each developer connecting directly to OpenAI or Anthropic with separate accounts and uncontrolled spending, Flock gives you one internal URL that everything points to. Flock then routes each request to a model running on your own machines, or to a paid cloud API, depending on how you configure it. You install Flock as a single binary on a Mac or Linux machine. It works with several local AI engines, including Ollama, vLLM, MLX-LM, and llama.cpp. Once running, you can load open-weight models such as Llama, Qwen, or DeepSeek onto your hardware. Flock can spread requests across multiple machines if you have more than one, and can split a model too large for any single machine across several machines using llama.cpp-RPC sharding. For team management, Flock lets you issue per-user API keys and set daily request quotas per user. Every request goes into an audit log so you can see who sent what and when. A built-in web dashboard at the same port shows activity and lets you manage keys without touching any config files. Connecting existing tools is straightforward because Flock speaks the same API format as OpenAI and Anthropic. Tools like Cursor, Claude Code, Aider, or any Python or JavaScript SDK can point to Flock instead of the real vendor URL, with no changes needed in the tool itself. If a request is for a model only available from a cloud vendor, Flock proxies it through. If your local hardware can handle it, the request stays in-house and costs nothing in API fees. Getting started takes about three commands: install the binary, install a local engine such as Ollama, then run flock up with a model name. A built-in diagnostic command called flock doctor inspects your hardware and tells you exactly what to install if anything is missing.