⚡ What is AI Productivity Flow?
AI Productivity Flow is a desktop AI system engineered to operate seamlessly alongside the apps you already use. Instead of copying and pasting text into separate browser windows or siloed AI chat services, Flow stays quietly accessible across your entire operating system.
Select any text, notes, or documentation on screen, or tap a global shortcut to speak naturally:
- 🎬 Video Flow — Generate visual explainer videos with narration and synchronized motion graphics.
- 🎧 Audio Flow — Read text aloud or listen to structured multi-depth audio summaries.
- 🎙️ Voice Flow — Dictate directly into any focused application with instant text injection.
📥 Downloads & Installation
Windows 10 & 11 (64-bit)
macOS 14 Sonoma or later (Apple Silicon only, M1 or newer)
The Mac disk image is about 271 MB. The zip archive is about 287 MB.
Installing on Mac
The Mac build is Apple Silicon only (M1 or newer); it does not run on Intel Macs. Download links above stay on the latest stable release. On Mac you get all three: Video Flow, Audio Flow and Voice Flow.
- Open the DMG and drag the app to Applications.
- The app is not signed yet, so the first time you open it macOS blocks it. On macOS 15 and later, right-click → Open no longer gets past this. Instead:
- Open the app once. When macOS says it can't verify the developer, click Done.
- Go to System Settings → Privacy & Security and scroll down to the message about AI Productivity Flow.
- Click Open Anyway, then confirm (enter your password if asked).
- Allow Microphone when asked.
- For hotkeys and auto-paste, turn on the app in System Settings → Privacy & Security → Accessibility and Input Monitoring, then reopen the app.
- The default dictation shortcut on Mac is Cmd + Option.
After updating on Mac: when you install a new version, macOS may quietly stop honouring the permissions you gave the old one, so hotkeys or auto-paste stop working. Go to System Settings → Privacy & Security → Accessibility, select AI Productivity Flow, remove it with the minus (−) button, then add it again with + (or switch it off and on). Do the same under Input Monitoring, then restart the app.
🌟 The Three Pillars of Flow
1. 🎬 Video Flow — Turn Selected Text into Visual Explainer Videos
Video Flow transforms dense articles, technical notes, or documentation into structured, animated MP4 explainer videos with synchronized voiceover narration.
Platform note: Video Flow works on Windows and macOS (Apple Silicon).
Source Text / Notes
│
▼
AI Planning & Concept Extraction (Gemini / OpenAI / Claude / Local)
│
▼
Creative Director & Scene Authoring (Key ideas, visual hierarchy, layout)
│
▼
Controlled Rendering Pipeline (Kinetic typography, diagrammatic cards, transitions)
│
▼
Finished Explainer Video (.mp4 saved to Downloads + in-app player)
- Multi-Engine Rendering Pipeline: AI understands and plans scene intent; Flow's Creative Director shapes pacing, layout, and visual rhythm; controlled local rendering compiles the final video.
- Google NotebookLM Integration: Directly connect NotebookLM research notebooks for audio and visual synthesis.
- Visual Styles & Themes: Choose between distinct visual presentations (Dark Editorial, Minimal Technical, Kinetic Modern).
- Deterministic Fallback: Local deterministic visual layout engine ensures rendering completes reliably even if external APIs experience rate limits.
- Export & Control: Export high-definition MP4 videos directly to your system Downloads folder with playback controls and job management.
2. 🎧 Audio Flow — Spoken Readers & Multi-Depth Audio Summaries
Audio Flow gives you two dedicated ways to absorb written content without eye strain:
- How to start it: select text with the mouse in any app, then click the small Listen button that appears next to it (or the floating bar) and pick Read or Summary. On Mac you can also press Control+Option+L to read the current selection aloud; press it again to stop. On Mac, reading your selection needs the Accessibility permission.
- Read Mode (Spoken Explainer): An intelligent reader designed to articulate selected paragraphs, articles, or documentation naturally. Rather than robotic text playback, it speaks with natural human cadence, clear phrasing, and synchronized playback.
- Summary Mode (Multi-Depth Audio Summaries):
- Short: Fast high-level briefs for quick updates.
- Balanced: Structured conceptual overviews capturing key arguments and takeaways.
- Deep Dive: Thorough breakdown analyzing context, evidence, and implications for long documents.
- Integrated Audio Player: Mini on-screen and full-window player with playback rate controls (0.75x to 2.0x), waveform progress scrubbing, and instant MP3 download.
- Voice Ecosystem: The default voice is Microsoft Edge's online text-to-speech (free, no account required). It runs online, so the text being read is sent to Microsoft. You can also connect Cartesia, ElevenLabs, OpenAI TTS, or Google Cloud TTS, or use local system speech.
3. 🎙️ Voice Flow — System-Wide Voice Dictation & Text Injection
Voice Flow lets you speak naturally and injects clean, formatted text directly into whatever application is currently focused—Slack, Word, VS Code, Notion, browser inputs, or terminal windows.
- Local-First Dictation: Bundled local Faster-Whisper engine runs completely on-device across CPU or GPU without requiring internet access.
- Downloadable Streaming Models: Optional download of NVIDIA Nemotron Speech Streaming English 0.6B and Nemotron 3.5 ASR Streaming 0.6B for real-time streaming speech recognition.
- Cloud STT Providers: Connect Deepgram, Groq Whisper, OpenAI Whisper, or Gemini STT for instant cloud-accelerated transcription.
- On-Device & Cloud AI Polish:
- Downloadable on-device polish models (S1-mini by Superwhisper, Liquid LFM 2.5 350M) via llama.cpp for private text cleanup.
- Windows AI Text Rewriter where supported by hardware.
- Optional cloud LLMs (Gemini, Claude, OpenAI, Groq, Nebius Token Factory, Ollama) for smart cleanup, conversational rewriting, and prompt formulation.
- Voice Commands: Commands such as "make this an email" or "summarize this" need a cloud polish model (for example Gemini with your own API key). Local polish (S1-mini, LFM) cleans up your dictation but does not run commands.
- "Hey Voice Flow": This works inside a normal dictation, not as an always-on listener. Press your dictation hotkey, then say something like "Hey Voice Flow, make this an email..." The app only listens while you are dictating.
- Personal Dictionary & Snippets: Add custom acronyms, jargon, and snippet abbreviations that expand automatically with cross-process memory.
🔒 Privacy & Security Architecture
- Local-First by Design: Voice Flow speech recognition (Faster-Whisper, Nemotron) and local polish (S1-mini, LFM) run on your device, and that audio and text never leave your computer.
- When the cloud is used: only for cloud polish or cloud speech providers you set up with your own key, online text-to-speech voices (including the default Microsoft Edge voice for Audio Flow and Video Flow, which sends the text being read to Microsoft), and NotebookLM features.
- Explicit Cloud Routing (BYOK): When you select a cloud provider, only the text or audio required for that specific request is transmitted to that provider. No central tracking intermediary is used.
- Local Storage: Dictation history, audio recordings, and provider credentials reside locally in SQLite databases inside
~/.voice_flow/.
- Encrypted Backups: The
.flowvault export/import format secures account profiles, preferences, and custom dictionaries using AES-256-GCM encryption.
- Loopback Desktop API: The internal GUI HTTP and WebSocket server binds strictly to
127.0.0.1:8991 with Origin, Host, and port verification to prevent cross-origin browser access.
Telemetry & Crash Reporting
AI Productivity Flow is strictly privacy-first:
- On by default, with an opt-out: Crash reports are on for every install, including existing users who never changed the setting. Open Settings → System → About & Updates → Privacy and choose Turn off. That takes effect immediately, with no restart. An explicit opt-out stays off.
- Zero Audio / Transcript Exposure: Crash reports never contain microphone audio, speech transcripts, dictation text, clipboard contents, or request payloads.
- Strict Privacy Scrubbing: Stack traces and error messages pass through a local privacy scrubber before leaving your machine:
- Personal username paths (e.g.
/Users/<username>/... and C:\Users\<username>\...) are scrubbed and replaced with [REDACTED].
- API keys and secrets (
OPENAI_*, ANTHROPIC_*, GROQ_*, GEMINI_*, DEEPGRAM_*, SENTRY_*, AWS_*, etc.) are stripped.
- Performance tracing and user profiling are completely disabled (0% sample rate).
- How to Verify or Change: The Privacy row under Settings → System → About & Updates turns crash reports off or back on. You can also run
voice-flow --test-crash-reporting to test your current preference.
⌨️ Global Shortcuts
| Platform | Default Shortcut | Action |
|---|
| Windows | Ctrl + Win | Start / Stop voice dictation into active input |
| Windows | Alt + Space (Optional) | Configurable secondary dictation trigger |
| Windows | Middle Mouse Click (Optional) | Push-to-talk with mouse button |
| macOS | Cmd + Option | Start / Stop voice dictation into active input |
🛠️ Development & Building from Source
To set up a local development environment or contribute:
# Clone the repository
git clone https://github.com/uzerkayat2004-gif/AI-Productivity-Flow.git
cd AI-Productivity-Flow
# Create and activate a virtual environment
python -m venv .venv
# Windows:
.\.venv\Scripts\activate
# macOS / Linux:
source .venv/bin/activate
# Install editable package with test dependencies
pip install -e .
pip install pytest pytest-asyncio pytest-timeout cryptography
# Run the test suite
pytest tests/test_platform_layer.py tests/test_runtime_compatibility.py tests/test_version_consistency.py
For full details on project architecture, code conventions, and packaging pipelines, see CONTRIBUTING.md.
📄 License
AI Productivity Flow is open-source software licensed under the Apache-2.0 License.
Third-party component notices and licenses are documented in release/THIRD_PARTY_NOTICES.txt.
