lilbee
A single local executable that runs and manages AI models across every GPU on a machine and turns them into a citation-backed search engine over a user's own files, notes, code and the web.
lilbee is described as an entire local AI stack in one executable: it runs and manages chat, embedding, vision and rerank models spread across every GPU available, and puts them to work as a conversational search engine over everything indexed, with every answer citing the exact file and line it came from. It crawls websites into its library, can launch coding agents against local models, and exposes cited answers to any MCP-aware agent, all reachable as a terminal application, a command-line tool, an MCP server, an HTTP API or a Python library.
It is private by default, sending no telemetry and needing no account, and only contacts a cloud provider when an API key is explicitly configured for it. It has its own model manager built on llama.cpp, covering the architectures behind most GGUF models on Hugging Face, so it does not require Ollama or LM Studio, though it can also work alongside an existing installation of either. It is currently beta software distributed as PyPI pre-releases.