Press a key, talk the way you think, and LittleWhisper types the cleaned‑up version into any app on your Mac. Fillers gone, punctuation fixed, tone just right.
Any update on the design review? Want to lock it before Thursday.
faster than typing — most people speak 150+ wpm
from releasing the key to text on screen, with streaming
custom editor modes, dictionary rules and history
every feature, no license, no account, no cap
Fn, Globe or any shortcut you like. Works from any app — LittleWhisper lives in your menu bar.
Ramble, restart, say “um.” Audio streams to the transcriber while you speak, so nothing waits for you to finish.
Your editor mode cleans it up and LittleWhisper types it right where your cursor is — and keeps a copy on your clipboard.
A mode is an AI editor with its own instructions. Switch on the fly from the recording pill, or build your own for emails, notes, prompts or code comments.
“Um so basically the client wants to push the deadline to like the 15th, and uh we need to tell the design team, oh and the budget's still the same.”
LittleWhisper handles the whole loop — capture, clean up, insert, review — so you can trust it with anything you'd normally type.
Highlight text in any app, press the shortcut, pick a mode. LittleWhisper rewrites it in place — no dictation needed.
Audio is transcribed while you're still talking, so stopping is the last step — not the start of a wait.
See exactly what the AI changed in every transcription, Word‑style. Undo the editor with one click.
Every transcription is saved on your Mac. Search it, flag it, or re‑run it through a different mode.
I wanted to follow up on the design review…
✨ CleanHi Sam, thanks for sending the invoice over…
🖋️ ProfessionalTeach it names, brands and jargon it keeps mishearing.
If a transcription fails, the recording is saved to History so you can retry instead of re‑dictating.
Auto‑detects what you're speaking, or pin a language for the best accuracy.
Music and videos pause while you talk and resume when you stop — only if they were playing.
Lists, bold and line breaks paste with formatting into apps that support it.
Built‑in, USB or AirPods. If your Mac switches microphones mid‑sentence, recording carries on.
Also: focus or minimal recording overlay · custom shortcuts · light & dark mode · launch at login · sound cues · automatic updates
Nobody types three paragraphs of context into Cursor or Claude. Everybody will say them. Talking gives AI tools the detail they actually need.
Use the latest models with your own keys, run everything on your Mac, or skip the setup entirely. Mix and match transcription and editing.
Your audio and text go straight to the provider you choose. No middleman, no markup.
Download a model once and transcribe and edit without an internet connection. Nothing leaves your machine.
Rather not deal with API keys? Buy a credit pack, sign in with Apple or email, and go. Credits never expire.
Download and start talking. An account is only needed if you choose Credits.
Recordings are discarded once transcribed. A failed one stays on your Mac for up to 7 days so you can retry.
Transcriptions, modes and dictionary live on your Mac. Delete any of it whenever you like.
With on‑device models your audio and text never leave your Mac. Usage stats are anonymous and never include audio, text or keys.
No subscription, no license, no feature gating. You only ever pay a provider directly — or nothing at all with on‑device models.
| LittleWhisper | Typical dictation apps | |
|---|---|---|
| Pricing | ||
| Price | $0 · optional Credits | $8–15/mo or one‑time license |
| Account required | No | Usually |
| Features behind a paywall | None | Common |
| Transcription | ||
| Streaming while you speak | Yes | Some |
| Providers | OpenAI, Deepgram, Groq, on‑device | 1–2, fixed |
| Works fully offline | Yes | Some |
| AI editing | ||
| Custom modes with your own prompts | Unlimited | Rare |
| Editor providers | OpenAI, Anthropic, Google, xAI, on‑device | 0–2 |
| Per‑mode model choice | Yes | No |
| Rewrite highlighted text (Edit Selection) | Yes | Rare |
| Track changes on every edit | Yes | No |
| Everyday | ||
| Custom dictionary | Yes | Some |
| Searchable history with reprocessing | Yes | Some |
| Retry failed recordings | Yes | Rare |
Yes. Every feature is free with no license, account or usage cap. You either bring your own API key (and pay that provider directly, usually a few cents a day), run models on your Mac for nothing, or buy optional Credits if you'd rather skip keys.
No. On‑device models need no key, and Credits need no key. If you already have an OpenAI, Anthropic, Google, xAI, Deepgram or Groq key, paste it in and you're set — it's stored encrypted on your Mac.
Anything with a text field — Slack, Mail, Messages, Notion, Google Docs, Cursor, VS Code, ChatGPT, Claude, Teams and more. It types where your cursor is and always leaves a copy on the clipboard.
With streaming (OpenAI or Deepgram), text typically appears in under a second after you stop talking; we measured about 0.7 seconds on a 38‑second dictation. Editing with a fast model like GPT‑6 Luna added about 0.8 seconds in the same test.
No. Audio is discarded once it's transcribed. If a transcription fails, that recording is kept on your Mac for up to 7 days so you can retry, then deleted.
Yes — download an on‑device Whisper model for transcription and a Gemma model for editing, and LittleWhisper works with no internet connection at all.
macOS 14 Sonoma or later on Apple Silicon or Intel. LittleWhisper asks for microphone access and Accessibility permission (so it can type into other apps).