langolier

Langolier is a local knowledge studio that indexes documents, code, videos, and voice notes, allowing users to ask questions and receive answers with citations. It operates entirely on the local machine unless configured to use a cloud API. The system indexes content the user already owns and answers from it, using SQLite for retrieval with BM25 for words and vectors for meaning, fused with RRF, and optionally reranked by a judge model. When no corpus content supports an answer, it abstains rather than inventing one.

The tool supports ingestion of many file types including PDF, DOCX, PPTX, XLSX, ODT, ODP, ODS, Markdown, text, CSV, JSON, notebooks, source files, and subtitles. It can transcribe video and audio through whisper.cpp, including voice memos. Watch folders can be set to local or mounted NAS paths, polled on a timer, with sync or hoover modes. Named assistant profiles can be created with their own mission, scope, engine, caps, and theme, and exported as .langolier bundles with standalone launchers. A Telegram bridge uses long polling without opening a port, and a local API provides one JSON endpoint on loopback for Shortcuts, Siri, and local agents. On macOS 26 and newer, an Apple engine can answer from the system's model without downloading anything.

Answers are grounded: every factual sentence carries a numbered citation, and chips under the answer open the exact passage that backed it. Strict mode answers only from retrieved passages with inline citations. The footer shows which memory mode and which model produced the answer.