Drop a recording into the window and a minute later it is text. Not just a transcript: a summary with decisions and tasks, a readable version and a verbatim text with timecodes. Your Mac does the work on its Neural Engine — no account, no subscription, no cloud.
6.8 MB · Apple Silicon · macOS 14+ · free, MIT
Nothing to set up: the app lives in the menu bar and waits for you to hand it a recording.
Drag a file into the window, drop it onto the menu bar icon, or forward a voice message to your own Telegram bot. Any format: mp3, m4a, wav, mp4, mkv, webm, ogg.
An hour of audio takes about a minute. You can close the window — the queue keeps going, and the menu bar icon shows how much is left.
A recording opens on its summary: what it was about, what was decided, who does what. The verbatim text is one click away, for when you need the exact words.
All three are produced once, during processing, and all three are kept. Switching between them is instant — nothing is recomputed and nothing is sent anywhere again.
Opens first. What people almost always want is the answer to “what was this about”, not forty minutes of text.
The same speech without fillers, slips and false starts, broken into paragraphs. The version you can forward to someone.
The raw model output with per-word timecodes — for quoting and for checking against the audio.
The window opens from the menu bar — though on an ordinary day you never need it.
Forward a voice message to the bot and get the text back. No static IP, no port forwarding, no domain: the bot lives inside the app and works as long as your Mac does.
You can’t type in someone’s @username — until their first message the bot doesn’t know it. They write to the bot, you get a prompt with “Allow” and “Decline”.
Whoever writes first is a candidate, not the owner: someone could have found the bot by name before you did. The app shows who it was and asks.
A stranger is told their request went to the owner; if declined, that access wasn’t granted. Otherwise they’d keep sending voice messages into the void.
Transcription services ask you to upload the recording of your meeting. This one doesn’t ask.
The recording is transcribed on your Mac. It is not sent to any server — ours or anyone else’s. The app doesn’t even keep the audio: the source file stays where it was.
The network is used twice: once to download the recognition model, and every few hours to ask for the latest version number. The summary is a separate, optional feature.
The source is open under the MIT licence: what the app does with your recordings is visible line by line, not taken on the developer’s word.
| Chronicler | Cloud transcription | Built into macOS | |
|---|---|---|---|
| Recording leaves the Mac | never | every one | for some languages |
| Works offline | yes | no | partly |
| Price | free | per minute or monthly | free |
| Summary with tasks | yes, on your own API key | varies | no |
| Speaker separation | yes | usually | no |
| Export | Word, Markdown, HTML, PDF, SRT, VTT, JSON | varies | plain text |
| Recording length | unlimited | by plan | short only |
No account, no installer: open and drag. Two minutes, and most of that is downloading the model.
Double-click the downloaded file, then drag the app into the Applications folder.
On first launch it downloads the recognition model — 461 MB, once. The system then spends about half a minute compiling it for your machine.
Any file: mp3, m4a, wav, mp4, mkv, webm, ogg. Or straight onto the menu bar icon — no need to open the window at all.
The app is signed but not notarised by Apple — that is a paid process and hasn’t been done yet. It takes two clicks to open, and only the first time: right-click the app → “Open”, then “Open” again in the warning dialog.
A double-click won’t do here: the system offers only “Cancel”. After that the app opens the ordinary way.