Molebie AI

Molebie AI is a self‑hosted assistant that runs entirely on a user’s hardware, keeping all data local and private. It combines voice interaction, image understanding, document memory, and web search without relying on cloud services. Users can talk to the assistant using wake‑word activation and speech‑to‑text powered by Whisper, and receive spoken responses generated with Kokoro TTS.

The system supports three “thinking modes” that let users choose between instant answers, chain‑of‑thought reasoning, or deeper extended reasoning with larger token budgets. Visual input is accepted via drag‑and‑drop or paste, and textual documents such as PDFs, DOCX, TXT, and Markdown can be uploaded for retrieval‑augmented generation using a hybrid vector and BM25 search with cross‑encoder reranking. Real‑time web results are fetched through a self‑hosted SearXNG instance, with LLM‑driven intent classification and source citations.

All persistent data is stored in a local SQLite database, offering multi‑user isolation and full ownership. The platform can be deployed on macOS, Linux, or Windows (WSL2) with a single‑command install script, and its components—web app, API, LLM inference, search, and text‑to‑speech—can run on a single machine or be distributed across multiple devices according to hardware capacity.