Frequently asked questions

What is FIA?

Forgenity Interview Assistant is a Windows desktop app that listens to the other party in a call, transcribes the speech locally, detects questions, and streams short answer hints into a small overlay — grounded in documents you choose yourself.

How does the answer generation work?

You point FIA at a folder of your own documents (project notes, CV, glossary). When a question is detected, FIA retrieves the most relevant passages locally and sends them — together with the question — to an OpenAI-compatible LLM endpoint (BgGPT by default). The model's reply streams into the overlay as bullet-point hints.

Does my audio leave the computer?

No. Speech-to-text runs on your machine using a local Whisper model. Only the recognized text (the question plus your retrieved notes) is sent to the LLM endpoint you configured. If you use a fully local model endpoint, nothing leaves the machine at all.

Will the overlay show up when I share my screen?

The overlay window is excluded from OS-level screen capture on Windows 10 2004+ — it does not appear in Teams/Meet/Zoom screen sharing, OBS display capture, or PrintScreen. It also stays out of the taskbar and Alt-Tab. Note: it cannot hide from a second camera filming your screen or hardware-level capture.

Does it work with phone calls?

Yes — via Windows Phone Link. Pair your phone over Bluetooth, take the call through your PC, and the caller's voice is captured through the same audio pipeline as a video meeting. See the Download page for setup notes.

What does it cost?

The app is free to download and use. You supply your own LLM API key — BgGPT's free tier works out of the box, and any OpenAI-compatible endpoint can be configured.

Which languages are supported?

Speech recognition uses Whisper, which supports Bulgarian, English, and ~90 other languages. The response language of the hints is configurable — pick the language in settings and the model replies in it.

What hardware do I need?

Windows 11 (or Windows 10 2004+). A microphone for your side of the conversation; the other party is captured via system audio (loopback), so no special setup is needed for online calls. For offline Whisper models a reasonably modern CPU is recommended.

Can it answer general knowledge questions?

Yes — technical terms and definitions are answered from the model's general knowledge even when they are not in your notes. Project-specific claims stay grounded in your retrieved documents.

Is this the same as Cluely?

FIA shares the genre — a real-time assistant that helps during calls — but it is built around your own documents, runs its audio pipeline locally, works with any OpenAI-compatible model, and is free. It is developed as part of the FORGENITY project family.