Name: Rekounts
Author: Rekreate Digital

Talk into anywhere

If you can type there, you can talk there. No extension, no copy-paste dance — it types where your cursor is.

For vibe coders

Vibe code without typing. Speak the prompt.

Good prompts are long — the context, the constraints, the why. Hold the hotkey and just say it. Rekounts types your words straight into Cursor, Claude Code, Copilot, Codex, Gemini — or whatever AI has your cursor.

Any tool, no plugin. It's keystrokes at your cursor — editors, terminals, browser chats. If a human can type there, you can talk there.
The context makes it in. Most people speak several times faster than they type, so the details you'd trim from a typed prompt survive the spoken one.
Your prompts stay yours. Dictation runs locally — Rekounts uploads nothing. What reaches an AI is only what you choose to send it.

AI assistant — your editor

›Refactor the auth middleware to return a 401 JSON body instead of throwing, retry once on token refresh, and update the tests — keep the logging exactly as it is.

Spoken in one take. Typed by Rekounts.

How it works

One hotkey. That's the manual.

Hold Ctrl + Win

Anywhere in Windows. A small charcoal pill above your taskbar brightens — it's listening. Don't like the combo? Rebind it in Settings.

Speak naturally

A Whisper speech model transcribes on your own machine. Your audio stays in memory. Nothing is sent anywhere.

Release. It types.

Cleaned-up text lands where your cursor is — email, chat, doc, code editor. Fillers stripped, sentences capitalized, your clipboard left alone.

Anywhere you type

Sounds good — we ship Thursday. I’ll send the notes around by lunch.

Hold Ctrl+Win to dictate

Prefer hands-free? Double-tap the hotkey and keep talking; tap once to stop. A forgotten recording stops itself after 10 minutes — with a warning first.

Privacy

Your voice never leaves your machine.

Not “we don't sell your data.” Not “encrypted in transit.” There is no transit. Speech recognition runs entirely on your computer, and the code is public — you don't have to take our word for any of this.

Audio stays in RAM

Recordings are never written to disk — no .wav files, no temp files. Audio lives in memory while you speak, goes straight to the local model, and is dropped.

Zero telemetry

The app contains no analytics code of any kind. No crash reporters, no usage pings, no “improve the product” uploads. Nothing phones home.

Checkable, not just claimed

Every promise on this page can be read in the source on GitHub — including the ones above. Yours to read, change and share.

The only two times it touches the network

verified in docs/privacy.md

Downloading a speech model — once per model. ~486 MB for the default, straight from the project's own release servers. Every file is verified against a SHA-256 hash before it's used. After that, it works with the network unplugged.
Check for Updates — when you click it. One request to the public GitHub API. It only runs on its own if you switch on the optional automatic check, which is off until you do.

Help, Send Feedback and clicking an update notification aren't in that list on purpose: they hand a page to your browser or a message to your mail client — already written, and not yet sent. Rekounts makes no request of its own, and never sees the answer.

Bytes of your speech sent anywhere, ever

This site keeps the same promise: no analytics, no trackers, no third-party requests. Fonts and code are served from this domain.

Features

Small app. Daily driver.

Works in every app

It types where your cursor is — no special editor, no extension. Paste mode borrows your clipboard and puts it straight back; keystroke mode types the characters, and borrows the clipboard only for long dictations, which arrive mangled if typed out.

Personal dictionary

Teach it names and jargon it keeps mishearing, with optional “sounds like” spellings. Words bias the recognizer; mishearings are auto-corrected.

Insights dashboard

Words today, this week and all time; average words-per-minute; daily streaks and a 21-day chart. Computed locally from your own history.

Cleans up as it goes

Strips the ums, capitalizes sentences, fixes punctuation spacing, removes repeated words and comma-marked hedges — light, predictable, local.

A sticky note that listens

Right-click the tray icon for a floating Scratchpad that's already listening. Dictate into it, format it, close it — your note is still there tomorrow.

Free forever

GPL-3.0 licensed. No trial that expires, no word limit, no “Pro” tier holding features hostage. If it's in the app, it's yours.

The Hub

Everything you dictate, on your disk, at your fingertips.

Searchable history, insights and your dictionary live in a single local SQLite file. Clear one entry, clear everything, or turn history off entirely — it's your data.

Illustration of the Hub's Insights page — same layout and palette as the real app.

Compare

An honest comparison.

Wispr Flow and Letterly are polished products with cloud AI rewriting and mobile apps — if that's what you need, they're worth paying for. Rekounts is the option for when you want dictation that is free, open and provably private.

Feature	Rekounts	Wispr Flow	Letterly
Price	Free, forever	Free tier, then Pro at $15/mo ($12/mo billed yearly)	Free to start, then Pro at $19.99/mo or $69.99/yr (US App Store)
Where your speech is processed	On your machine	Cloud	Cloud
Open source	Yes — GPL-3.0	No	No
Account required	None	Yes	Yes
Works offline	Fully, after one model download	No	Partly — records offline
AI rewriting & tone	Light local cleanup	Advanced cloud AI edits	Cloud AI rewrites & styles
Platforms	Windows (macOS from source)	Mac, Windows, iPhone, Android	iOS, Android, Mac, Windows, Web

Competitor details as of July 2026, per their public sites; Letterly prices its subscription in-app, so its US App Store listing is shown. Check their sites for current plans.

Install

Running in two minutes.

01
Download & run the installer

The button below gets you the installer. No administrator rights, no UAC prompt — it installs for your account only. (Prefer no installer? A portable ZIP is on the releases page: unzip it anywhere and run Rekounts.exe.)
02
Click through the warning

Windows SmartScreen will warn you because the app isn't code-signed yet. Click More info → Run anyway. Once. (Or build from source yourself — it's open.)
03
First launch fetches the model

A one-time ~486 MB speech-model download from the project's own release host, SHA-256-verified. Then the pill appears above your taskbar — and from here on it works fully offline.

Download for Windows

On macOS? It runs from source today — experimental, and not yet confirmed on real Mac hardware. There's no Mac download yet.

FAQ

Fair questions.

Why does Windows warn me when I run it?

The “Windows protected your PC” SmartScreen notice appears because the app isn't code-signed — a signing certificate costs money and this is a free project. Click More info → Run anyway; you only see it once. The entire source is public if you'd rather verify or build it yourself.

How big is the download?

The app folder is roughly 350 MB — it bundles Python, Qt and the speech engine so you don't install anything. On first launch it downloads the default speech model once (~486 MB). Optional sizes: base (~148 MB, fastest) and medium (~1.5 GB, most accurate) — each downloads only if you pick it.

Where do the speech models come from?

From the project's own release space on GitHub — the one place the app trusts for models. Every download is verified against a SHA-256 hash recorded in the app's source before it is used, and interrupted downloads resume.

Does it work offline?

Yes. After the one-time model download, transcription is fully local and works with the network unplugged. The only other request the app makes is “Check for Updates”. “Help” and “Send Feedback” make no request at all — they open a page in your browser, or a message in your mail client, for you to send yourself.

Windows only?

The download is Windows 10/11. macOS can run it from source today — every Mac feature is written, but it has never been run on real Mac hardware, so there is no Mac download yet rather than one that might quietly not work.

What languages does it support?

Auto-detect, English and Tagalog today, with a choice of model sizes trading speed for accuracy. Bigger models help — Tagalog in particular is noticeably better on medium.

How is it free? What's the catch?

No catch. Rekounts is GPL-3.0 free software built by Rekreate Digital, an agentic-AI studio, because we wanted a dictation tool that doesn't cost a subscription or upload our voice. There's no premium tier and no data to monetize — your history never leaves your machine.

Where does my data live?

In plain files in the app's data folder under %APPDATA% on your own disk: a readable JSON settings file and a SQLite database with your dictation history and dictionary. Nothing is encrypted because nothing leaves your machine — and you can clear or disable history entirely.