Press a hotkey, say what you want, and the words land where your cursor is. No cloud. No accounts. No telemetry. Sweet, fast, and stupid private.
Open the app once to pick languages and hotkeys. After that it lives in the tray; you'll never need the main window again unless you want to add a snippet or check yesterday's transcripts.
Words appear at your cursor. Five hotkeys cover the whole flow.
That's the whole UI. The pill does the heavy lifting.
Anything that takes text takes tellmebaby. Slack. Outlook. VS Code. Notion. Your terminal. Even sketchy little admin tools nobody else supports.
Six small skills, one tiny pill. Designed to feel like a friend hanging around in the corner, not a dashboard taking over your life.
Hold your hotkey, talk, release. The words land at your cursor. Slack, VS Code, Outlook, your terminal — anything that takes typed text takes voice now.
Highlight a paragraph, hit a hotkey, hear it. Same for the clipboard, or any chunk of text you paste in. Your computer talks back, in a real voice that doesn't sound robotic.
Select text, press the edit hotkey, say "translate to French" or "make this shorter and less formal." The selection rewrites in place — no copy-paste shuffling.
Got a name the recognizer keeps butchering? Add it. Snippets map voice phrases to fixed text — addresses, signatures, prompts you reuse. The longer you use it, the smarter it gets at sounding like you.
Clean up filler words in Slack but keep them verbatim in your terminal. Different apps get different treatment automatically — no settings to remember when you switch windows.
Parakeet TDT for English. Whisper Large v3 Turbo for everything else. INT8 on CPU, GPU acceleration where there is one. No "lightweight model" excuses for bad accuracy.
tellmebaby ships two ASR models that pick themselves automatically based on the languages you speak. Both rank near the top of the public Hugging Face Open ASR leaderboard. Both run in INT8 on a normal laptop CPU at multiples of real-time.
| Model | WER ↓ | Speed (RTF) ↑ | Languages | Size |
|---|---|---|---|---|
| Parakeet TDT 0.6B v3 | ~6.3% | ~30× | 25 EU (incl. English) | 660 MB |
| Cohere Transcribe 2B | ~5.4% | ~12× | 14 (EU + zh/ja/ko/vi/ar) | 1.7 GB |
| Whisper Large v3 Turbo | ~7.4% | ~4× | 99 | 1.0 GB |
| Whisper Large v3 (full) | ~7.7% | ~1× | 99 | 3.1 GB |
Onboarding is intentionally tiny: tell us which languages you speak. We auto-pick the right ASR model, download it in the background, and you're dictating before the install bar finishes filling.
Pick all that apply. We'll grab a model that handles them well — no model picker to fuss with later.
That's the whole onboarding.
Picked so they don't fight your browser, your editor, or Windows itself. Ctrl+Shift+Alt + a letter is essentially never claimed by anything else — so the chords stay yours.
all five are remappable in Settings → Activation
Hold Esc while the pill is active to throw away whatever you just said.
One small surface, bottom-center of your screen. It tells you what's happening — listening, transcribing, done — without stealing your attention. No window, no panel, no dashboard. Hover it for the hotkey hint or click to open the actual app.
Names of coworkers, internal tool names, that one client whose name nobody can spell. Add them once to your dictionary; the recognizer biases toward them forever. No model retraining, no cloud roundtrip.
The recognizer runs entirely on your CPU or GPU. No microphone audio is ever sent over the network. We say "100% local" and we mean it — pull your network cable mid-sentence and tellmebaby keeps working.
Cloud dictation is fast and accurate. So is tellmebaby — without sending your voice anywhere. Here's the actual difference.
"100% local" is easy to claim. Here's the technical reality so you can verify rather than take our word for it.
The recognizer is sherpa-onnx, an open-source ONNX Runtime wrapper that runs INT8 model weights on your CPU. No code we write touches a network socket during transcription. Open Wireshark; watch nothing happen.
Every release is signed with an ed25519 key (separate from Authenticode) and the public half is embedded in the app. The auto-updater refuses to install anything that doesn't verify against it. You can't be tricked into installing a malicious "update."
SHA256 of every release is published right next to the download button. certutil -hashfile setup.exe SHA256 on Windows or shasum -a 256 on Mac/Linux confirms what you got matches what we shipped — no MITM, no tampered binaries.
Both default models — Parakeet TDT v3 (NVIDIA) and Cohere Transcribe (Cohere Labs) — ship under Apache 2.0. You can fork tellmebaby, swap the models, redistribute. No vendor can pull the rug.
Recordings live as .wav in ~/.tellmebaby/recordings/. Transcripts live in a SQLite DB you can open with any client. Settings are JSON. Nothing's compressed, nothing's hidden, nothing's "cloud-synced for your convenience." Delete the folder; reset complete.
No paid tier planned. No telemetry to monetize. If maintenance ever shifts to paid, current installs keep working at the price they were installed at. The version you have now is the version you have forever.
No cloud accounts. No telemetry. No analytics beacons. The only network traffic is downloading speech models the first time you pick a language — and even those are open-weight files you can delete and re-fetch any time. Your voice is yours.
If you've ever finished an email and realized your hands are tired, tellmebaby is for you. If you haven't, it's for you anyway.
You think faster than you type. Talk through your draft, then clean it up with the keyboard. tellmebaby gets the messy first version out so editing is the only work left.
Commit messages, doc strings, Slack replies, code reviews. The 40% of dev time that's English instead of code is where dictation pays off — your hands stay on the keyboard, your voice does the typing.
Eating lunch. Holding the baby. Pet on the lap. Whatever's happening, you can still get a paragraph out. tellmebaby doesn't care if your fingers are sticky.
The hotkey is "any modifier you can mash with one finger." Custom dictionary, fast cancel, everything works one-handed. Edit Mode means you don't even need to retype to revise.
Run the installer, pick the languages you speak, and start dictating in under five minutes. Updates are automatic and signed.
Download v0.2.4 for Windows%USERPROFILE%\.tellmebaby — recordings as .wav, transcripts in a SQLite database, settings as JSON. Nothing's compressed, nothing's encrypted, nothing's hidden. You can browse it all directly in File Explorer. Deleting that folder is a complete reset.
Win + letter at the shell level (so we never see Win+R, etc.), and browsers eat Ctrl+Shift+letter shortcuts (Ctrl+Shift+R reloads, Ctrl+Shift+V pastes plain text). Ctrl+Shift+Alt + letter is essentially never claimed by anything else, so the chords stay yours no matter which app is in front. Awkward to type — but they're usually held briefly on one hand. Re-bind to whatever you prefer in Settings → Activation.