The speech model downloads once on first start (a few hundred MB, depending on the size you choose — small or medium).
Settings → LLM → API key. BgGPT is preconfigured; any OpenAI-compatible base URL works.
Point FIA at the directory with your documents. Each subfolder can hold
a бележки за интервю.md file — use the included template.
The hint language is independent of the speech language — listen in Bulgarian, get hints in English, or vice versa.
First public build: local Whisper transcription, WASAPI loopback capture, question gate, document-grounded hints (local retrieval over your notes), private always-on-top overlay, streaming OpenAI-compatible LLM support, startup version check.