Your voice, your machine.

Press a hotkey. Speak naturally, or circle anything on your screen. FlowState transcribes, formats, and pastes directly into your workflow without ever touching the cloud.

v1.0.0-beta · ~1 GB · Windows 10/11
RECORDING · VERTICAL LED METER
HUD PREVIEW
Explore

What it does

🎙️

Speak, Don't Type

Press your global hotkey anywhere in Windows. Speak naturally. Your words are transcribed by Whisper's large-v3-turbo model running entirely on your GPU, with lightweight CPU fallback if needed.

Ctrl + Shift + Space GLOBAL HOTKEY
🧠

AI-Cleaned Text

Raw speech becomes polished prose. A local 1.5B parameter language model restructures run-on sentences into clean lists, proper punctuation, and natural formatting.

"um so check the logs and like restart the server"
✔ Check logs & restart server
🎯

Point at Your Screen

Circle anything on your display while speaking. FlowState captures an annotated screenshot and links it to exactly what you were saying. Perfect for code reviews and bug reports.

⌨️

Paste Anywhere

Formatted text lands directly in whatever app you were using: Slack, VS Code, Gmail, Notion, Discord. Your original clipboard is seamlessly restored after.

VS Code Slack Notion Gmail Obsidian
🔒

Zero Cloud Leak

Both AI models run 100% locally. No audio recordings, no transcripts, and no metadata ever touch a remote server. Not even crash analytics.

0 BYTES SENT OUT

How It Works

From natural voice and screen gesture to polished Markdown or email copy in three friction-free steps.

01
RECORD

Trigger & Speak

Hotkey triggers your microphone. Speak at your natural pace with no robotic commands, pauses, or special cadences needed.

🎙️ Audio Captured
02
PROCESS

Local AI Clean

Whisper converts speech to tokens. A local Qwen model removes filler words, formats code, bullet points, and grammar in milliseconds.

⚡ Qwen2.5 1.5B Restructuring
03
PASTE

Instant Injection

Formatted text is pasted directly into your active window. Screenshots attached automatically. Clipboard contents preserved.

📋 Pasted into Active Window

Under the hood

Built for developers and privacy-conscious users. Transparent, auditable, and completely offline.

Component Technology Implementation Details
Speech Engine faster-whisper Whisper large-v3-turbo with CUDA acceleration; fallback to base.en on CPU
Text Brain llama-cpp-python Qwen2.5-1.5B quantized GGUF, local prompt restructuring & grammar cleanup
Screen Capture mss + Pillow Ultra-low latency screenshot capture, gesture reticle detection, visual grounding
Desktop Shell PySide6 High-DPI neo-brutalist Qt interface, custom floating HUD & visual feedback
System Hooks pynput + pywin32 Global hotkey hooks, active window focus management, lossless clipboard restoration
0
Automated Tests Passing
0
Cloud Requests or Telemetry
0%
Open Source (MIT Licensed)
"Your thoughts deserve to stay yours."
🔒 Zero telemetry 🚫 No cloud servers 💾 100% Local execution

Ready to try it?

One installer. No sign-ups. No accounts. Just speak.

↓ Download FlowState for Windows
v1.0.0-beta · MIT License · Windows 10/11 64-bit