Voice2Note
Local-First & Air-Gap Capable

Privacy Center & Technical Transparency

Voice2Note is engineered from the ground up to respect your confidentiality. Spoken audio recordings and personal reflections should never be harvested for third-party model training.

100% On-Device AI

Speech-to-text, summarization, task extraction, and RAG vector search execute on your computer. Audio and prompts are never sent to external AI APIs.

Local Storage Only

All audio files are saved in data/audio/ and metadata in data/voice2note.db. No cloud S3 buckets or remote databases.

Zero Telemetry

No analytics SDKs, tracking pixels, or third-party cookies. The application functions cleanly even on an isolated network with zero egress.

Model Architecture & Licenses

AI Processing: Local
Speech Recognition
Whisper (whisper-tiny.en)
Runtime: In-Process ONNX Runtime (Transformers.js)
License: MIT / Apache 2.0
Reasoning & Extraction
Llama 3.2 (3B Instruct)
Runtime: Ollama Local Inference (localhost:11434)
License: Meta Llama 3.2 Community License
Vector Embeddings
all-MiniLM-L6-v2 (384-d)
Runtime: In-Process ONNX Feature Extraction
License: Apache 2.0
Vector Database
SQLite Float32Array Index
Runtime: In-Memory / Disk Cosine Similarity
License: Public Domain (SQLite)

How to Verify Offline Operation

You can test and demonstrate that Voice2Note works completely offline without network access:

  1. Ensure Ollama is running locally with ollama serve.
  2. Disable your Wi-Fi or turn off your machine's network interface completely.
  3. Navigate to Audio Studio, record a voice note, and click Save.
  4. Observe Whisper transcription, Llama 3.2 extraction, and vector search execute on-device with zero network requests.