Press a hotkey, dictate naturally, polished text appears in any app.
Ramble, use filler words, correct yourself mid-sentence. FreeFlow turns messy speech into clean writing and injects it wherever your cursor is: your messaging app, your editor, your coding agent, the terminal, email, anything.
It is open source, so you have the freedom to customize it any way you want. You deploy it on a server private to you, so your audio and data flow through infrastructure you control.
One deployment serves your entire team with no per-seat fees.
In this demo, you'll hear rambling speech with filler words and corrections. Watch what appears at the cursor.
freeflow-voice.mp4
Install the macOS app with Homebrew or download the DMG directly.
brew install build-trust/freeflow/freeflow
On first launch, FreeFlow deploys the code in this repo to a private server for you in Autonomy. You don't need to know anything about servers or cloud infrastructure. The whole setup takes about two minutes. Start dictating right away, and invite your team so they can use the same private server.
It feels instantaneous. Two thirds of dictations complete in under 0.6 seconds (benchmarks):
Audio streams to your private server over a persistent WebSocket while you speak. The server forwards it to a realtime model, which transcribes incrementally. By the time you release the key, the transcript is already done. A skip heuristic bypasses the polish step entirely for clean transcripts (roughly 40% of dictations). When polish is needed, a fast model handles it in about 0.4 seconds.
Two independent WebSocket connections are kept warm: a primary that streams audio during recording, and a standby that races the primary with the full audio buffer when you release the key. If both WebSockets fail, an HTTP batch fallback catches it. Whichever path finishes first wins.
Everything is in this repo: the app, the server, the deployment configuration. Change the models, rewrite the prompts, add a language, or fork the whole thing. Your team's audio, transcripts, and context data flow through your private FreeFlow Server deployed in Autonomy. This gives you privacy and scale.
One container handles your entire team with no per-seat fees. In our concurrent users benchmark, 50 people dictating simultaneously produced sub-second median latency with zero failures. Each dictation only occupies the server for a few seconds, so 50 concurrent slots supports hundreds of users in practice. The economics improve as you add people because the infrastructure cost is small and fixed.
FreeFlow is designed to be taken apart and reassembled. Swap the speech model, rewrite the polish prompt, add a language, or change how text is injected. See CUSTOMIZE.md.
Jump in, we'd love your help.
The single most useful contribution right now is mic compatibility data. FreeFlow works well with built-in mics and AirPods, but every USB mic, headset, and audio interface is different. The app's "Contribute Mic Data" menu item generates a one-click diagnostic report that can help us improve accuracy of dictation for everyone.
Want to add or improve support for a language? Here's how. Found an app where injection breaks? Open an issue. Code contributions and pull requests are welcome too. DEVELOP.md has the build and test guide.
