Transcribes both sides of a call in real time, surfaces relevant notes from your knowledge base as you speak, and writes session summaries when you're done. No data leaves your computer.
Features
Captures mic and system audio simultaneously via Windows WASAPI. Both sides of the conversation transcribed as you speak, with speaker labels.
While the meeting runs, WinOpenOats searches your notes and surfaces the most relevant documents — you always have the right context on hand.
When the session ends, an LLM summarizes the transcript into structured notes automatically. Past sessions are saved and browsable.
Use Ollama for LLM + embeddings and Whisper or Parakeet for transcription. Nothing leaves your machine — optionally zero cloud APIs required.
WinOpenOats optionally operates entirely offline. Choose Ollama for the language model,
Whisper or Parakeet for transcription, and local embeddings — no API keys, no data sent anywhere.
Screen share protection via Windows SetWindowDisplayAffinity keeps the assistant
hidden from screen recordings.
Transcription Models
| Model | Size | Notes |
|---|---|---|
| Parakeet TDT v2 | ~600 MB | English, fast, NeMo backend |
| Parakeet TDT 1.1B | ~1.1 GB | Multilingual, NeMo backend |
| Whisper Base | ~142 MB | Good for low-resource machines |
| Whisper Small | ~244 MB | Balanced speed / accuracy |
| Whisper Large v3 Turbo Recommended | ~800 MB | Best accuracy / speed balance |
| Whisper Large v3 | ~1.5 GB | Maximum accuracy |
Setup
Requires Python 3.12 and a microphone. Run the onboarding wizard to pick your model and configure API keys.
For cloud mode: add your OpenRouter API key for LLM and Voyage AI key for embeddings in Settings. For fully local mode: install Ollama and pick local models — no keys needed.
Drop Markdown files into your knowledge base directory. WinOpenOats indexes them and will surface the most relevant ones during every call.
Free, open source. Windows 10/11. Python 3.12.