FAQ

All packages are on GitHub Releases: Windows provides MSI installers and portable ZIPs, Android provides direct APK downloads, and macOS has a notarized build. Install, open it, and follow the pairing guide. The download link always points at the newest version; iOS is in public beta on TestFlight.

On the same LAN, prefer LAN direct: nothing leaves your network and pictures keep their original quality. Across networks (phone on cellular, away from home) the cloud relay carries it, or run your own relay. Both do the same things; only the channel differs.

Self-hosted means the server runs on your own machine, with STT / LLM and a runtime you provide: speech recognition can use FunASR / Whisper / sherpa-onnx and similar, translate and organize use any OpenAI-compatible LLM with your own key, and the recommended default combination is written down in the deployment docs. The server source is public under AGPL-3.0, in the GitHub repository linked in the header.

Only Cloud relay needs a console account: register or sign in, bind your PC and phone on the devices page, then pair by QR code or pairing code. Self-hosted needs no sign-in, has no telemetry, and does not need the console at all.

Realtime: code comments, replies, filling in forms — say it and it is on screen. Translate: speak in your language and produce English or another target language, good for mail and documents. Organize: long spoken passages are tidied into cleaner prose before they land, good for dictating at length.

It depends who is reading. For an AI (a chat box, a coding assistant), use Realtime and send it as-is — understanding speech is what they are good at. For a human, or when structure matters, use Organize so it is tidied into cleaner prose before it lands. When the other side does not read your language, Translate swaps it for English, Japanese, Spanish and so on.

Yes. Over LAN direct, the original picture goes through; sent over the cloud relay it is compressed first — for original quality, use LAN direct. Boxes that accept pasted images get it directly, otherwise paste manually on the PC.

Yes. Choose Quick notes while speaking and it stays in the phone timeline; deliver it to any paired PC whenever you like. It goes to the machine you pick, not to whichever device happened to be connected at the time.

Sign up on the website and you are on Free straight away. Pro and Max are monthly subscriptions: open the billing page in your account, choose a plan and pay by card. You can cancel at any time, and the plan keeps working until the end of the period you have paid for.

The Refund Policy page, linked in the footer of every page, has the detail. A new subscription carries a 14-day right of withdrawal: ask within that window and you get your money back. Cancelling is a different thing — it stops the next renewal, and the period you have already paid for runs to its end.

Both the PC app and the phone app can export logs from their settings. Issues are collected in one place — the GitHub repository linked in the header.

All three product apps (server, desktop, mobile) are AGPL, and the protocol is Apache-2.0. What stays closed is the cloud operations side: this site, the console and the billing gateway, the same split Plausible and Bitwarden use. The cloud sells convenience, never capability: no product feature will ever live only in closed code.

No. The mic runs only while you hold the talk button — release and it stops. No screen reading, no screenshots, no background capture. FlowMic also never registers your phone as a microphone device on the PC — Zoom cannot use it. LAN keeps audio and text on your network when your STT runs there too; point it at a cloud STT key of your own and audio goes to that provider.

FlowMic combines AI-assisted development with rigorous human architectural governance. Security-critical surfaces—injecting text into windows, device pairing authentication, and end-to-end cryptographic transport—are audited line-by-line by senior engineers with minimal dependencies. Core code is open source on GitHub for public review, and dependency metrics are documented in SECURITY.md.

Win+H uses your PC's mic, grabs focus, and stays on one machine. FlowMic makes your phone the mic — no fight over your PC's mic or focus — and is open source, self-hostable, engine-agnostic, and streams spoken Chinese into written English. Closed cross-screen IMEs exist; FlowMic is the open, engine-neutral, privacy-first take. The boundaries are just as plain: it is not remote control, it never presses Enter or Send for you, and your PC does not switch input methods.

Windows 10 or 11, or a Mac on Apple silicon, plus an Android phone or an iPhone / iPad through the TestFlight beta. On one network the two connect directly; across networks the cloud relay carries it (or your own relay). On the cloud relay, transcription is managed for you; self-hosted, you bring STT / LLM and the runtime — FunASR / Whisper and similar, or your own cloud key.

For light use — the odd message, a prompt now and then, notes taken while you are away from the desk — Free covers it. If you dictate for most of the day you will use it up before the month ends, and that is what Pro and Max are for; you can start or cancel either one from your account. Running your own relay and your own engines is free and is not metered.