Privacy & trust
The short version: your audio and your transcript stay on your computer. Here is the whole story, in plain language.
What is captured
Only what you explicitly choose, in one of three modes — and the mode picker on the front page is the only thing that decides which.
This computer's audio: when you click Record, your browser shows its own picker and you choose one tab (or window, or screen). Live Transcribe receives that source's audio — and a tiny throwaway video signal the browser insists on sending along, which is shrunk to a single pixel and never shown, saved, or used.
Microphone: the microphone is captured only after you pick that mode and your browser asks you and you allow it — never in the other mode, never on page load. By default your device's own automatic gain and noise reduction are left on, so quiet and distant voices are picked up — all of it processing that runs on your device like everything else. A visible “raw signal” toggle turns both off for close-up recording in a quiet room. Echo cancellation is always off (it is phone-call processing that smears room recordings). One honest limit: on a phone, a phone call takes the microphone over, so no app — this one included — can transcribe a phone's own calls. Microphone mode is for rooms you are in, or a second device beside a speakerphone.
Audio file: a recording already on your device, read only when you pick it in the file dialog. It is decoded and transcribed on your machine like everything else, and the file itself is never copied anywhere.
What the app works out about a recording
After a recording, the app shows “Key excerpts” and “By the numbers”. Both are computed on your machine and both are mechanical: the excerpts are sentences quoted word for word from your own transcript, chosen by counting which sentences share the most vocabulary with the rest — and the numbers (words per minute, questions asked, sentences with numbers) are counted, with each line pointing at the exact moment it came from. Nothing here is written by an AI, nothing is a judgment, and none of it is a summary. The optional “tidy the transcript” switch only ever removes filler words and stuttered repeats — it never adds or rewords anything, and switching it off shows the exact words again.
Where the transcription comes from
The speech-recognition engine is a piece of software that runs inside your browser, on your computer's own processor. The first time you record, it is downloaded once from this site — the dialog that asks first always shows the exact size, which depends on the quality level you pick — and kept in your browser's storage, so later visits skip the download. Every file is checked against a known fingerprint before it is used. No audio is ever sent anywhere to be transcribed — the transcription happens entirely on your machine, which is the whole point of the download. The one exception is described under Server recorder below, and it exists only when this site's owner has switched it on: a recording the server itself made can then be transcribed on the server, by an engine inside it, and never anywhere else.
What is stored on your computer
- The engine files, in your browser's storage, so they download only once.
- Transcript checkpoints, saved in your browser as the words are recognized — this is what lets a transcript survive a crashed tab or a dead laptop battery — and, if you opened a server transcript here for review, this browser's copy of it, marked as the server's, kept until you delete it from the review screen or until the recording is deleted in its section.
- The audio recording, kept temporarily in your browser's storage during a session so you can choose “Save Audio” afterwards. It becomes a file of yours only if you save it.
Everything above lives in your browser's storage for this site, on your disk, under your control. By default nothing is stored on a server — this site's server hands out the app and the engine files and keeps none of your text. Two features can change that, both off unless switched on deliberately, and both described in full below: the Server recorder and the Transcript bridge.
What is sent over the network
By default, nothing of yours. The page and the engine files come down from this site; no audio, no transcript, no analytics, and no telemetry go up, and the browser's own security policy for this site blocks talking to any other server. If you are technical, you can verify this yourself: open your browser's developer tools on the Network tab during a recording and watch nothing leave. The nuances below are the complete list of exceptions, and the last one is the only thing here that can send your words anywhere.
One nuance, stated fully: the “My calls” page can show upcoming earnings dates from a public calendar feed. That feed is fetched by this site's server from a data provider (Alpha Vantage), at most once a day, and the request is identical no matter who visits or what calls anyone follows — so the provider learns nothing about you, and your browser still talks only to this site. Your calls list itself lives in your browser's storage and never leaves your device.
A second, narrower nuance: a call's Prep button (and the official-transcript viewer behind it) asks this site's server for that one company's earnings history or a past call's published transcript. Those lookups are symbol-specific — the data provider sees which ticker the server asked about (though not who asked, or anything else). They happen only when you click, are cached on the server for a day, and nothing about your recordings or transcripts rides along. After a recording, the fact check's “Check against the … call” button is the same kind of lookup: it asks the server for that one company's earnings history (if the Prep card was not opened) and one prior call's published transcript, only when you click it, and nothing from your recording rides along. If that trade is not worth it to you, don't click Prep or that button — nothing else uses these lookups.
The third nuance is the Server recorder, and it is the one feature where this site's server touches audio at all. When its owner has switched it on, you can hand the server a direct media link (a replay's playlist, say) and it downloads that stream to its own disk — only the link you gave it, nothing from your device. When its owner has also switched on server transcription — a separate switch, and on this site the owner turned it on on 5 September 2026; the recording's card says whether a server transcript exists — the server transcribes that recording itself with a speech-recognition engine that runs inside the server on its own processor: the same kind of on-machine engine your browser uses, only on the server's machine instead of yours, and it sends nothing anywhere to do it — no audio and no text leaves that server. Said exactly, because it is the boundary that matters: the server transcribes only recordings the server itself made — from a link you handed it, or from a capture its owner scheduled — and never anything captured on your device. Your tab audio, your microphone and your imported files are transcribed in your browser, as the section above says, and nothing about this feature changes that. With that switch off, a server recording is transcribed only when you click Transcribe here, in your browser. The recording and any transcript the server made of it rest on the server until you delete the recording, which removes both; the transcript is also filed where the Transcript bridge (below) files transcripts, when this server has that store at all, whether or not you have switched the bridge on in your browser — because it was made on the server and was never on your device. You can download the transcript from the recording's card, or open it here for review — this browser then keeps its own copy of those words, in the same storage as your transcripts, marked as the server's, and never sends that copy anywhere (the Transcript bridge below is not told about it, because the server filed its own copy when it made the transcript); the review's Delete transcript removes this browser's copy and nothing on the server, and says so. Or still click Transcribe here to transcribe the recording in your own browser instead. The access token you paste is a capability key for that server, kept in this browser's storage — it is not a password to anything else. If the section isn't visible, the feature is off and this paragraph describes nothing.
One sentence of that paragraph needs widening, because until recently it was true only while somebody was watching: when this site's owner has armed a scheduled capture, the server can start recording at a set time with no browser open at all — nobody present, nothing on your screen. Said exactly: at that time, or later within a limit the owner sets, so a server that was restarting at the appointed moment starts once it is back rather than not at all. It records only a media link the owner placed on the server by hand; it never goes looking for one, it never records anything from your device, and a recording started that way is transcribed the same way as any other server recording: on the server, by the engine inside it, with nothing sent anywhere, when the owner has switched server transcription on — otherwise in a browser, when someone clicks Transcribe here — and never anything from your device.
The fourth nuance is the Transcript bridge, and it is the only feature that sends what was said off this device. Switched on, each paragraph is copied to this site's server as it is committed, so a long call survives the tab closing and can be read while it is still running. Said plainly, because it is the one claim on this page that a feature can reverse: with the bridge on, your transcript text is stored on the server — not the audio, which never leaves your browser, but the words. It is off until you switch it on yourself, in the Transcript bridge section on the front page: that section takes an access token of the bridge's own, and turns the bridge off again in one click. The Server recorder's token does not turn it on, and neither does anything else. Turning it on during a session also sends what that session has already committed — the words so far, not just the words to come — and timestamped notes you type travel with the transcript while it is on, and are re-sent when the session ends if they failed to land, the same way the words are. While the bridge is on, that section says so and shows what is still waiting to be sent, so you are never guessing; an armed bridge always shows its section, even when the server cannot be reached. Switching it off stops anything further leaving; what was already sent stays on the server until the transcript is deleted — and deleting is in your hands: after a recording (or on reopening a recovered one), the review screen's Delete transcript button removes this browser's copy — the words and any notes you typed on them — and, for a transcript made in this browser with the bridge on, asks the server to delete its copy of that transcript too, telling you once the server has confirmed. (For a server transcript you opened here, the same button removes only this browser's copy and says so; the server's copies go with the recording, in its section.) If the bridge is off — or the server cannot confirm at that moment — the browser's copy is still removed at once, and the request to delete the server's copy is kept in this browser and sent the next time the bridge is on here; the note the button leaves and the bridge section both say while one is waiting. One carve-out, stated plainly: that kept request lives in this browser's storage, so clearing this site's data before it is sent abandons it — then removing the server's copy is the server owner's act. If the section is not on the page at all, no bridge token is stored here and this server offered nowhere to put transcripts — nothing leaves; and if it is there but you have not switched it on, this paragraph still describes nothing and the Network tab check above shows nothing leaving. One exception, stated under Server recorder above: transcripts the server makes of its own recordings are filed in that same store when the store exists, because those words were never on your device — and they leave when that recording is deleted in its section, not through this button, which reaches only transcripts made in this browser.
How to wipe everything
Because everything is in your browser's storage, your browser's own controls remove it all: click the padlock/tune icon next to the address, open Site settings, and choose Delete data (in Chrome: Settings → Privacy → Site settings → View permissions and data stored across sites → this site → Delete). That removes the cached engine, all transcript checkpoints (a copy of a server transcript you opened here included), and any unsaved audio. Files you saved with the Save buttons are ordinary files of yours — delete them like any other file. One carve-out, stated plainly: what already left for the server is not in your browser's storage, so this does not touch it — a Server recorder recording, and any transcript the server made of it, stay until you delete the recording in its section, and anything the Transcript bridge sent stays on the server until that transcript is deleted (for a transcript made in this browser, the review screen's Delete transcript button reaches the server's copy too, as the bridge section above describes) or until it is removed on the server itself. Clearing this site's data also discards a delete request the bridge was still waiting to send.
What does not exist yet
Honesty about the edges: there is no sharing feature, no account, and no cloud sync. If any of that is ever added, it will be a visible choice — not a quiet change to where your words go. (Microphone mode arrived exactly that way: as a visible choice on the front page, never a default.) The server does not transcribe anything uploaded from a device, offers no retry button for a failed server transcript, and transcribes English only.