Say it. It types.
Hold a key and speak. Let go, and your words appear wherever your cursor is, in any app: English, Spanish, or both in the same sentence.12
Free: 15 minutes a week. No card needed.4
Tersa Cloud is in alpha: these are early builds.
DictationReady
Hold to try it
Hold to talk. Double-tap for hands-free. Press and hold, then let go. The words are an example, not a recording.
MicrophoneMic open
The dot turns red only while the microphone is open.1
This weekFree
15:00left
of 15:00 · resets Monday
Each dictation can run 2 minutes.4
RecentExamples, not recordings
Dile a Marco que the build passed y que ya puede hacer el merge.
MessagesPasted
Could you send me the deck?
MailPasted
MessagesListeningPasted · Cloud · Soniox
Hold. Speak. Let go.
Mic open
Hold your key and speak.
Right Ctrl, or any key you choose. The words appear as you talk, in a small bar at the bottom of the screen. Double-tap the key to keep talking hands-free.1
Pasted
Let go.
The text lands where your cursor is, in the app that had focus when you pressed: an email, a chat, a document, a code editor.1
History
Nothing is lost.
If the text can’t be pasted, it’s copied for you. Offline, the recording waits and is transcribed when you’re back. History keeps your dictations on your computer, for as long as you choose.1
English, Spanish, or both at once.
Set the language to Automatic and speak the way you do: a sentence can start in one language and finish in the other. Names and jargon you add to Your words go with each dictation, so Tersa can get them right.12
Language you speak
HistoryExamples, not recordings
Can you send me the deck antes de la reunión? I’ll review it tonight.
PastedCloud · Soniox
Dile a Marco que the build passed y que ya puede hacer el merge.
PastedCloud · Soniox
As spoken, Clean or Polished.
Free includes As spoken. Clean and Polished require Pro, including the active seven-day trial. These buttons show examples, not a live dictation service.12
Cleanup
Keeps every word, with punctuation.
Removes fillers and fixes punctuation.
Also smooths grammar and wording. A language model suggests the edits, and each one is checked before it is inserted.
- Heard
- um, so can you, like, send me the deck
- Inserted
- um, so can you, like, send me the deckSo can you send me the deck?Could you send me the deck?
Nobody spoke? Nothing billed.
Minutes count the audio of each dictation, to the millisecond. A take with no speech in it is never sent to a speech engine and uses no minutes. Neither does a take that fails.3
| Length | Recent | Minutes used |
|---|---|---|
| 0:04.2 | Can you send me the deck antes de la reunión? I’ll review it tonight. | 0:04.2 |
| 0:01.6 | No speech | 0:00.0 |
Where your voice goes.
History and Your words stay on your computer. Tersa’s servers do not store your audio or your text: both are discarded when a dictation ends. The privacy policy has the details.25
Your computer
While you hold the key, the app streams your audio to Tersa.
Tersa’s service
It checks your plan and minutes, and passes the audio to one speech engine once you start speaking.
One speech engine
Soniox, or Groq if Soniox is unavailable. With Polished, the text also goes to a language model: Cerebras, or Groq as a fallback.
Back to your app
The text comes back and is pasted. History shows which engine handled each dictation.
Start free. Try Pro for 7 days.
Prices in US dollars. Taxes are calculated at checkout.
Pro
$12a month
Paid: up to 15 hours a month · All three modes4
Trial: 60 total minutes over 7 days, no card. Paid Pro is a separate purchase.
Sources
- 1Tersa Cloud desktop app and approved Mac release scope, 2026-10-23: hold to talk, hands-free, local History and Your words. Public downloads offer Apple Silicon Mac; legacy Windows/Linux download and update URLs remain available at their frozen releases.
- 2Tersa Cloud API README, 2026-09-28, and the production deployment of 2026-10-05: the speech route (Soniox, then Groq), English and Spanish in one take, and the Polished edits checked by a validator before they leave the service.
- 3Tersa Cloud API README, 2026-09-28, “Nobody spoke, nothing billed” and metering: no audio reaches a speech engine before speech starts, a take with no speech or a failed take costs nothing, and usage is counted in milliseconds of audio.
- 4Approved plans, 2026-10-23: Free has 15 minutes/week, two-minute takes and As spoken. Paid Pro costs $12/month or $120/year, with 15 hours/month, ten-minute takes and all three modes. The explicit no-card seven-day Pro trial has 60 total minutes, ten-minute takes and no automatic conversion or charge.
- 5Tersa Cloud’s production deployment, 2026-10-05: the diagnostics store is switched off there, so no dictation’s audio or text is stored on Tersa’s servers.