I built this because I wanted the AI summary features mainly to summarize the action items discussed in a meeting, one on one convo or talking to yourself. The main logic here is that everything works offline.
Fission runs entirely on-device:
Transcription: Uses Vosk for offline STT.
Summarization: Runs a quantized Llama model locally to summarize the transcript. Uses Qwen 0.6b llm model.
Stack: React Native / Expo.
It’s open source (GPLv3). I’d love feedback on how the inference performs on older devices. I'm currently working on better options for transcribing.