How It Works

Understand how Prismical captures audio, transcribes locally, and generates structured notes.

Architecture Overview

Prismical runs as a native desktop app that captures system audio and processes it with AI:

Audio Capture — Prismical listens to your system audio output. No bot joins your meeting — it captures whatever your computer plays or records through the microphone.
Transcription — Audio is processed by local AI models (Whisper, Parakeet) running on your device. Optionally, you can use cloud providers (Deepgram, OpenAI, Assembly AI) with your own API keys.
AI Processing — Language models analyze the transcript to generate summaries, extract action items, and identify key decisions.
Storage — Notes, transcripts, and summaries are stored locally on your machine.

Transcription Engine

Prismical supports multiple speech recognition backends:

Local Models

Whisper — OpenAI's open-source model, runs entirely on-device
Parakeet — NVIDIA's ASR model, optimized for speed on Apple Silicon and CUDA GPUs
Speaker diarization (who said what)
Multiple languages and accents
Technical vocabulary handling

Cloud Providers (BYOK)

Deepgram, OpenAI, Assembly AI — bring your own API keys
Audio is sent to the provider for processing (only when you opt in)
Useful for older hardware or when you prefer cloud accuracy

AI Summarization

After transcription, AI generates:

Summary — Concise overview of what was discussed
Key Decisions — Important decisions made
Action Items — Tasks with assignees extracted from conversation
Follow-ups — Topics that need further discussion

Voice Notes & Quick Notes

The same pipeline works for non-meeting content:

Voice notes — Talk into your mic, AI transcribes and structures
Typed notes — Type messy thoughts, AI formats and organizes
Both go through the same AI processing to produce clean, structured output

All processing happens locally by default. Audio is processed in memory and never stored as a file. Only the resulting text (transcript, summary, action items) is persisted to local storage on your machine.

If you use cloud transcription providers, audio is sent to those providers for processing according to their privacy policies. You control this choice in Settings.