Voice2Note

Settings & Local AI Configuration

Configure model parameters, demo dataset, storage, and the primary friend persona.

Interactive Demo Dataset

Demo Mode Available

Instantly populate Voice2Note with a realistic 6-note dataset (engineering syncs, product pricing, marathon training, home renovation) containing pre-indexed vector embeddings and extracted tasks to test search and grounded RAG immediately.

Target Friend Persona

Build for a Friend

Configure the primary user persona. The structured intelligence extraction tailors summaries and task prioritizations to this workflow.

Technical Model Transparency

Privacy Center
Speech Model
whisper-tiny.en

License: MIT License

Language Model (LLM)
Llama 3.2 (3B)

License: Meta Llama Community License

Embedding Model
all-MiniLM-L6-v2 (384-d)

License: Apache 2.0

Inference Runtime & Vector Store
ONNX Node + Ollama + SQLite

In-process float32 cosine similarity

Local AI Service Offline

Ollama could not be contacted at http://127.0.0.1:11434. To enable intelligence extraction and RAG queries, run:

ollama run llama3.2

Storage & Security Policy

Zero Cloud Egress: All audio files are saved in data/audio/ and database records are stored in data/voice2note.db.

Strict Path & Magic-Byte Validation: Uploaded audio files reject traversal attempts (`..`, `/`) and verify non-executable magic headers before disk writes.