b0yko/tern is a tool for Apple Silicon that searches podcast and video libraries locally. It combines Whisper speech recognition, Apple Vision OCR, and SigLIP-2 visual search into a single ranked list. The repository uses FastAPI and vanilla JS with Tauri, and is written in Python. FACTS records no public releases.
The tool turns the words in speech, the text on screen, and scene images into separately searchable information. According to the README, it processes audio with Whisper and Silero VAD, reads on-screen text with an Apple Vision helper, and finds scenes with SigLIP-2 embeddings. The results it produces come together in one list. After an editor for trimming segments, you can export to video, audio, subtitle, and table formats, and FCPXML export is also supported. The README lists 539 tests, including pipeline and API checks along with a license check.
That said, the test count does not mean search accuracy or processing speed, and FACTS provides no performance figures to compare. The tool targets Apple Silicon, and whether the same features work in other environments cannot be confirmed from the given information. The license is marked NOASSERTION and the repository description states that all rights are reserved, so the reuse conditions are also unclear. It is therefore close to a local search tool whose scope and terms of use need to be checked.