Getting an AssemblyAI API key
AssemblyAI is the alternative speech engine — it does the same job as Soniox, turning what you say into words. Pick it if you already have an account there, or if you would simply rather not depend on Soniox. Vito works the same either way.
About five minutes, and an email address. AssemblyAI hands you a key as soon as you have an account, which makes this the quickest of the three to set up.
Finding your key
- Sign up at assemblyai.com. AssemblyAI advertises $50 in free credits on signup, no credit card required — which is a great deal of dictation, and the reason this is the easiest of the three to try. You can have Vito transcribing before you have given anyone a payment detail.
- Open the dashboard. Unlike most providers, AssemblyAI shows your API key on the dashboard itself rather than hiding it behind a creation step. Copy it from there.
- Add a payment method when the free credit runs out. Until you do, transcription stops once the credit is used up. While you are in the billing settings, set a spending limit or a usage alert — see below.
- Paste the key into Vito. Settings → Speech recognition, choose AssemblyAI, paste. Vito verifies it before you rely on it.
Prefer to read it from the source?
AssemblyAI's own documentation lives at assemblyai.com/docs. If the dashboard doesn't look like this description, theirs is the authority — this page will always lag a redesign.
What it costs
Nothing, until the free credit runs out. After that AssemblyAI bills per hour of audio, and the rate depends on which model you are on — at the time of writing their listed prices run from $0.15 per hour on Universal-2 up to $0.45 per hour for real-time on their newest Pro model. Soniox, by comparison, lists $0.10 to $0.12.
So the trade is a real one and worth knowing before you choose: AssemblyAI is the easier start, Soniox is the cheaper run. Vito's rule of thumb of €0.15 per hour of speech is based on the Soniox setup; on AssemblyAI expect somewhat more, depending on the model.
Language coverage differs by model too — their newest Pro model lists 18 languages while Universal-2 lists 99, against Soniox's 60+. Which of those is "more" depends entirely on which model you end up using and whether your language is on the shorter list, so check before assuming either way.
Checked on the AssemblyAI pricing page and their supported languages documentation in July 2026. Theirs are the numbers that count.
Looking after your key
This part is the same whichever provider you picked, and it is worth two minutes. None of it is paperwork: a key that gets out is somebody else spending your money.
Treat it like a password, because that is what it is
Anyone holding your key can make requests that you pay for. Don't paste it into a chat, a forum post, a bug report or a screenshot, and never commit it to a repository — bots scan public code for keys and find them within minutes.
Set a spending limit before you start using it
All three providers let you cap what the account can spend in a month, or warn you at a threshold. This is the most effective protection you have and it takes a minute to set. With a limit in place, a mistake costs you the limit instead of the balance.
Keep a little credit on the account
These are prepaid services. Run the balance to zero and transcription simply stops working — usually with an unhelpful error. Topping up a small amount and setting a low-balance alert saves you diagnosing a problem that isn't one.
Most providers also offer an automatic top-up: when the balance drops below an amount you choose, they charge your card for another fixed amount. Worth switching on. It is the difference between dictation that always works and dictation that stops on a Tuesday morning because you forgot to log in somewhere three weeks ago.
Automatic top-up needs the spending limit, not instead of it
The two belong together. A top-up that refills on its own, with no ceiling above it, is a bucket with the tap left running: whatever goes wrong keeps being paid for. Set the monthly limit first, then let the top-up keep you under it.
Make a key that only Vito uses
If you already use the same provider for something else, create a separate key rather than reusing one. It costs nothing, it tells you from the provider's usage page what Vito actually spends, and if you ever need to revoke it you won't break anything else in the process.
If it leaks, revoke it
Deleting a key at the provider makes it useless immediately — there is no grace period to worry about and no way for anyone to keep using it. Create a fresh one, paste that into Vito, and carry on. Rotating a key is a two-minute job, so do it at the first suspicion rather than waiting for certainty.
The bill is yours
You are the account holder. The provider bills you directly, at their prices, for whatever the key is used for. Vito is not a middleman: it holds no account on your behalf, never sees your usage, and cannot cap, refund or dispute anything. Which also means the savings are yours — nobody is taking a cut — but so is the responsibility.
Where Vito keeps it
On your own machine, in Vito's settings, and it is sent only to the provider it belongs to. Nobody at Vito can read it. That is the point, and it has one consequence worth knowing: nobody can recover it for you either. Lose it and you make a new one.
If something doesn't work
Vito says the key is invalid
Because the key sits on a dashboard rather than in a copy-once dialog, the usual culprit is an incomplete selection — a character left behind at either end. Select it again, carefully, and paste it fresh.
It stopped after a while
Most likely the free credit ran out. Add a payment method, or top up, and it resumes.
Should I use this or Soniox?
Soniox is what Vito recommends, and it is what the €0.15 per hour figure is based on. AssemblyAI exists so you are not locked to one supplier. Neither choice is permanent: swapping is a matter of pasting a different key.
After this
That covers the listening. The tidying up — the part that turns a raw transcript into text you would actually send — is a separate service with its own key.