VATS — Versatile Audio Transcription & Summarization

Local-first audio/video transcription, speaker diarization, and AI-powered summarization with multi-provider support and GPU acceleration.

Features

• Whisper Transcription — OpenAI Whisper and faster-whisper backends with automatic model caching
• Speaker Diarization — pyannote.audio-based speaker identification
• 5 AI Providers — Ollama (local), Google Gemini, DeepSeek, OpenRouter, Z.ai
• Bulk Processing — Concurrent multi-file pipeline with multi-GPU support
• Document Summarization — Summarize TXT, MD, PDF, and DOCX files
• Performance Profiles — speed / balanced / quality presets
• Persistent Cache — SQLite-backed cache with LRU memory layer
• Prompt Templates — Configurable prompt library for different content types

Quick Start

1. Install

# Clone the repo
git clone https://github.com/dtembe/vats.git
cd vats

# Install (editable mode recommended)
pip install -e .

# Or use requirements.txt
pip install -r requirements.txt

2. Configure

cp .env.example .env
# Edit .env — set AI_MODEL, API keys, Whisper model, etc.

3. Run

Windows launcher (interactive menu):

vats_start.bat

Desktop UI (cross-platform):

cd desktop
npm install
npm run tauri:dev

See desktop/README.md for details.

CLI:

# Process a single file
vats process recording.mp4

# Transcribe only (no AI summary)
vats process recording.mp4 --no-summary

# Bulk process
vats bulk file1.mp4 file2.mp3 file3.wav

# Summarize a document
vats summarize report.pdf

# System status
vats status

# Cache management
vats cache stats
vats cache cleanup --force

As a Python module:

python -m vats process recording.mp4

AI Providers

Set AI_MODEL in .env to one of:

Provider	Value	API Key Required	Notes
Ollama	ollama	No	Local LLM, install Ollama
Google Gemini	gemini	Yes	GEMINI_API_KEY
DeepSeek	deepseek	Yes	DEEPSEEK_API_KEY
OpenRouter	openrouter	Yes	OPENROUTER_API_KEY — access Claude, GPT-4, etc.
Z.ai	zai	Yes	ZAI_API_KEY

Prompt Templates

Prompts live in prompts/ and are selected via SUMMARY_PROMPT_FILE in .env:

Template	Use Case
vats_summary_prompt_extract_wisdom.txt	General wisdom extraction (default)
vats_summary_prompt_meeting.txt	Meeting summaries with action items
vats_summary_prompt.txt	Quick executive summary
vats_summary_prompt_document.txt	Document analysis
vats_summary_prompt_video.txt	Video content analysis
vats_summary_prompt_technical_document.txt	Technical document analysis
vats_summary_prompt_generic_content.txt	Generic content summarization
vats_concall_summary_fabric.txt	Earnings / conference call analysis
vats_summarize_meeting_fabric.txt	Structured meeting fabric format

Performance Profiles

Set PERFORMANCE_PROFILE in .env:

Profile	Whisper Model	Segment Length	Workers	GPU Memory
speed	tiny	60s	4	95%
balanced	small	120s	3	90%
quality	medium	180s	2	85%

Project Structure

vats/
├── vats_start.bat          # Windows interactive launcher
├── pyproject.toml           # Python packaging
├── requirements.txt         # pip dependencies
├── .env.example             # Configuration template
├── prompts/                 # AI prompt templates
│   ├── vats_summary_prompt_extract_wisdom.txt
│   ├── vats_summary_prompt_meeting.txt
│   └── ...
├── desktop/                 # Cross-platform Desktop UI (Tauri + React)
│   ├── src/                 # React frontend
│   ├── src-tauri/           # Rust backend
│   └── README.md            # Desktop UI documentation
└── src/vats/
    ├── __init__.py
    ├── __main__.py          # python -m vats support
    ├── cli.py               # CLI entry point
    ├── app.py               # Main orchestrator
    ├── config.py            # Configuration & utilities
    ├── audio.py             # Audio extraction & segmentation
    ├── transcribe.py        # Whisper transcription backends
    ├── diarize.py           # Speaker diarization
    ├── summarize.py         # AI summarization (5 providers)
    ├── format.py            # Transcript & summary formatting
    ├── cache.py             # Persistent cache (SQLite + LRU)
    ├── bulk.py              # Bulk processing & GPU management
    └── docsummarize.py      # Document summarizer

Requirements

• Python 3.10+
• FFmpeg (in PATH)
• CUDA-capable GPU (optional, for acceleration)
• Ollama (optional, for local AI summarization)

Security

• Never commit .env — it contains API keys. The .gitignore already excludes it.
• Copy .env.example to .env and add your own keys.
• All AI API calls use HTTPS with certificate verification enabled.
• Credentials are loaded from environment variables only — no hardcoded secrets.

License

MIT — see LICENSE.