Lilbee, by Tobocop2, is an MCP server tool for local AI stacks, search engines, and model management. It is built with Python, using llama.cpp for its model engine and LanceDB for vector indexing. Lilbee offers self-contained binaries for Linux, macOS, and Windows, plus distribution via Homebrew, AUR, Nix, Docker, Flatpak, Snap, and Scoop. It can also be a Python library or run via a Node.js launcher. This tool focuses on local operation and privacy.

What does Lilbee connect to?

Lilbee manages local AI models, including chat, embedding, vision, and rerank, across available GPUs. It connects to a user's local files, code, and crawled web pages to power a search engine that provides cited answers. Optionally, Lilbee can integrate with existing Ollama or LM Studio instances. Users can also configure it to connect with cloud models, extending its capabilities beyond local resources.

Core features and tools

Lilbee exposes several callable tools for managing its functions. These include lilbee_search for queries, lilbee_add for content, and lilbee_crawl to index web pages. For status and control, it offers lilbee_crawl_status and lilbee_crawl_cancel. Model management is handled by lilbee_model_list and lilbee_catalog_browse. Other tools like lilbee_init, lilbee_remove, lilbee_list_documents, lilbee_export_dataset, lilbee_status, lilbee_sync, lilbee_settings_set, and lilbee_settings_reset provide system control.

Who is it for?

Lilbee targets developers, programmers using coding agents, and Obsidian users. It also serves individuals seeking a private, local AI search engine and model manager. Its design prioritizes local operation, ensuring privacy by sending no telemetry and requiring no account unless cloud models are configured. This is not the right fit if you require a service that inherently sends telemetry or demands cloud account integration for basic operation.

How to get started

For local deployments, the server authenticates using environment variables or system credentials. For remote HTTP-based MCP servers, access is granted through a standard OAuth authorization flow.

Use cases

• Local AI Search and RAG: Index personal documents, code, and crawled web content to receive cited answers to natural language queries.
• Coding Agent Integration: Provide a retrieval backend for MCP-aware coding agents, enabling access to cited snippets from indexed code and documentation.
• Local Model Management: Browse, download, and run various AI models (chat, embedding, vision, rerank) directly on local GPUs.

Lilbee provides a private, local AI stack for managing models and performing cited searches across personal data. Its support for various platforms and integration with MCP-aware agents makes it a versatile tool. The focus on local operation and privacy offers a secure environment for AI tasks. Start by using "lilbee_catalog_browse" to explore available AI models for local deployment.