Index — multi-source indexing

Entire YouTube channels, PDFs, DOCX, audio, images, WhatsApp, and meeting transcripts — all extracted and indexed locally, with no intermediary server.

Augment — local RAG

BM25 + FTS5 + CrossEncoder (and optional vector search) retrieve the most relevant excerpts from your base and deliver them to the model as context — without hallucinating beyond what was indexed.

Chat — sourced answers

Every answer cites title, date, and origin link. Runs offline with Ollama, or with Groq, OpenAI, Anthropic, and Gemini as optional providers.