Audio & Performance: - Add bytemuck for zero-copy f32→bytes conversion (O(1) cast) - Client backpressure throttling with THROTTLE_THRESHOLD_MS - Disable Whisper VAD for streaming (vad_filter=False) - Cache meeting_db_id in MeetingStreamState to reduce DB overhead - Adaptive partial cadence multiplier under congestion Diarization: - Auto-trigger offline diarization refinement (diarization_auto_refine setting) - Reduce streaming window: MAX_TURN_AGE 15→5min, MAX_TURN_COUNT 5000→1000 Testing: - Add 34 audio format conversion tests (test_audio_processing.py) - Fix pre-existing test failures: path prefixes (_mixins→mixins) - Fix test_sync_orchestration: error_code vs error_message Quality: - All checks pass: basedpyright (0/0/0), quality tests (90+28), grpc (678)
NoteFlow
NoteFlow is an intelligent meeting notetaker: local-first audio capture, navigable recall, and evidence-linked summaries. It uses a gRPC API for bidirectional audio streaming and transcription.
What's in this repo
- Python backend (gRPC server + domain services):
src/noteflow/ - Tauri + React desktop client (Vite):
client/
Prerequisites
- Python 3.12+
- Node.js 18+ (for the Tauri/React client)
- Rust toolchain (only if running the Tauri desktop app)
- Docker (optional, for containerized dev)
Local development
1) Backend server
python -m pip install -e ".[dev]"
python -m noteflow.grpc.server
Optional: run with autoreload (watches src/ and alembic.ini):
python scripts/dev_watch_server.py
By default the server listens on localhost:50051.
If you want PostgreSQL persistence, set NOTEFLOW_DATABASE_URL in your environment or .env (see example.env).
2) Tauri + React client
cd client
npm install
# Web (Vite dev server)
npm run dev
# Desktop (Tauri dev)
npm run tauri dev
The Tauri desktop app requires a working Rust toolchain.
Container-based development
The repository includes a compose.yaml with a server container (and a commented-out Postgres service).
Option A: Run the server in Docker, clients locally
- Create a
.envfile fromexample.envand set any needed settings. - Start the server container:
docker compose up -d server
The server will expose 50051 on the host; point your client at localhost:50051.
Option B: Enable PostgreSQL in Docker
compose.yaml includes a commented db service using pgvector/pgvector:pg15. To use it:
- Uncomment the
dbservice anddepends_on/environmentlines incompose.yaml. - Set
NOTEFLOW_DATABASE_URLto the container URL (example):
postgresql+asyncpg://noteflow:noteflow@db:5432/noteflow
- Start services:
docker compose up -d
Common commands
# Backend
pytest
pytest -m "not integration"
ruff check .
mypy src/noteflow
# Client (from client/)
npm run lint
npm run format
npm run test
Configuration
Runtime settings are read from .env and NOTEFLOW_ environment variables. See src/noteflow/config/settings.py and example.env.