How to ask for help

At a glance. How to ask for help and what to include (key, time, X-Request-Id) so we can answer fast.

One address and that's it: support@uttera.ai. A person answers.

There's no form, no ticket number, and no bot asking whether you've tried restarting. When the volume calls for it we'll build something with more machinery, and it will be said here; until then, an email arrives sooner and gets answered just the same.

If what you want to know is whether the service is down rather than whether it's just you, that's checked before writing: service status.

What to send us

This is almost always enough to work out what happened without having to ask you for anything else:

Item	Where it comes from
`X-Request-Id` — the most important one	A header on the response that failed. It identifies your exact request among all of the day's.
Approximate time and time zone	Your clock. Without the zone, a time is no use at all.
The endpoint	`/v1/audio/transcriptions`, `/v1/summarize`…
What you expected and what happened	You. The literal error, if there is one, better pasted than summarized.
Your customer identifier	It's on your account page. It is not a credential: knowing it grants access to nothing.

Never send us your API key. Not the whole thing, and not "the first few characters just in case". We don't need it for anything: with the X-Request-Id we find the request. If we ever receive a key in an email, we will revoke it and tell you.

And if the problem is with a specific audio file, you don't need to send it to get started. Almost everything is diagnosed from the request identifier. If we ever needed the audio, we'll ask for it and explain what for — which is the only way a recording of yours can end up in our hands for longer than the seconds it takes to process it.

How long we take

This is what we commit to. It is time to first response on business days, not time to resolution: some faults are fixed in ten minutes and others aren't, but knowing that somebody is looking at it shouldn't depend on luck.

Plan	First response
Free	As we can. No commitment — but it does get read.
Startup · Developer	2 business days
Professional	1 business day
Business	4 business hours
Enterprise	Direct contact, agreed in the contract

This is not the SLA. Your plan's SLA is a commitment on service availability and stands on its own. An email answered late doesn't consume SLA, and a service that's down isn't fixed by answering an email quickly.

If something is genuinely broken

If you think you've found a bug of ours — not a usage question — say so in the subject. We deal with it before anything else, and if it turns out you were right we'll publish it: in this documentation if it changes something written here, and in the changelog if it touches the service.

And if what you've found is a security problem, write it to the same address with [seguridad] in the subject and don't publish it in the meantime. We'll reply with what we know and with a timeframe.

Frequently asked questions

At a glance. Common questions and errors. For exact prices, limits and languages we link to their own page, so nothing goes stale here.

Getting started and account

How do I get an API key? Sign up at app.uttera.ai and generate it in your account. It travels in the Authorization: Bearer sk-... header.

Can I have several keys? Yes. Create one per integration (web, mobile, a connector…) so you can revoke one without touching the others.

How do I revoke or rotate a key? From your account: delete the old one and create a new one. Revocation is immediate.

Is there a free plan? Yes, with a monthly quota to try it out. What each plan includes is on Plans.

How do I upgrade or downgrade? From your account. On an upgrade the prorated difference is charged and the new plan applies immediately.

Are there annual plans? Yes, with a discount over monthly. See Plans.

How do I see how much I've used? Your account shows usage, remaining credit and a request history.

What happens when I run out of credit? You get alerts at 80% and 100%. Once it's gone, paid requests return 402 until you top up or upgrade.

Can I buy extra credit without changing plan? Yes, there are credit packs that add to your quota.

How do I pay? By card, and also with cryptocurrency (Bitcoin). The invoice includes your VAT/Tax ID if you set it in your account.

Errors and codes

401 — unauthorized. The key is missing or wrong. Check that you send Authorization: Bearer <your-key> and that the key is still active.

403 — forbidden. Your plan doesn't include that action. See what each plan includes on Plans and upgrade if you need it.

413 — too large. The audio or text exceeds the limit for that action. Shrink or shorten the file, or use async mode. Limits are in Reference.

402 — insufficient credits. Your monthly credit is used up. Buy a credit pack or upgrade.

429 — too many requests. You're going too fast. Retry with a growing wait (backoff) between calls.

500 / 503 — server error. Usually temporary. Retry; if it persists, write to us with the X-Request-Id.

504 on a long summary. The synchronous mode of /v1/summarize cuts off at 10 minutes and does not charge. For long audio use async mode, which has no such cap.

What is the X-Request-Id? A response header that identifies your exact request. Send it to us when asking for help and we'll find what happened without needing anything else.

Transcription

Which audio formats do you accept? wav, mp3, m4a, ogg, opus, flac, webm, aac and amr.

Is there a maximum audio size? Yes, and there is a duration cap in synchronous mode. For large or long files, use async mode. Detail in Reference.

Which languages do you transcribe? Many; the list per action is in Reference. If detection fails, pass the language explicitly.

My transcription came back empty or wrong. Usually audio with no speech/silence, a wrongly detected language (set language) or a corrupt file. Check the audio and retry.

Does it tell me who speaks and when? Yes, with diarization: it returns segments with speaker and timestamp.

Is it compatible with the OpenAI API? Yes. Point your OpenAI SDK at our endpoint and use the same names (e.g. whisper-1). See Integration.

Voice and cloning

Which voices and languages are there? A catalogue of voices per language; the list is in Studio.

Can I clone a voice? Yes: you give a short sample of the voice and the text to say. There is a character limit per request (see Reference).

Do I need permission to clone a voice? Yes. A voice is biometric data: only clone your own voice or one with explicit consent. Cloning a third party to impersonate them is illegal.

Does the generated audio carry a watermark? Yes, an inaudible watermark required by the EU AI Act (Art. 50.2). It cannot be disabled.

Translation

Do you translate text and voice? Both: to text, or translated and spoken.

Can I keep my own voice when translating? Yes: when translating voice you can keep the speaker's original voice instead of one from the catalogue.

Which translation languages are there? In Reference. Some codes have quirks (e.g. Norwegian or Chinese); the docs note them.

Summary and verifiable report

Can you summarize a recording? Yes: it transcribes and returns a structured summary (key points, data, to-dos, who spoke).

What is the verifiable report? A signed PDF with an evidence graph anyone can check: it links each statement to the moment in the audio. See Verifiable report.

How do I verify a report? With the verify action: it checks the signature (Ed25519) and that the PDF hasn't been modified.

Sounds and music

Do you generate sound effects? Yes, from a description (a door, rain, footsteps), and also sound scenes with several events mixed together.

And music? Yes, a piece from a description. The cost depends on length × quality steps; tune both.

What licence does generated music have? For your use under the terms; detail in the Sounds and music docs. It carries the Art. 50.2 watermark.

Pronunciation

Do you analyze pronunciation? Yes: you compare a recording with a reference sentence and it returns accuracy, expected vs heard phonemes and grouped errors. See Pronounce.

Privacy, legal and data

Do you train on my data? No. Your audio and text are not used for training nor reused. See Privacy.

Where is my data processed? In the EU, on our own infrastructure. See Privacy and Sustainability.

How long do you keep the audio? In async jobs the material is encrypted per job and deleted automatically after a short period. Detail in Privacy.

Do you have GDPR / a data processing agreement (DPA)? Yes. See Legal framework.

Can I ask you to delete my data? Yes, write to us. See Legal framework.

Integrations and connectors

Does it work with the OpenAI SDK? Yes: change the base_url to our endpoint and reuse your code. See Integration.

Do you have a connector for Claude / ChatGPT / Gemini? Yes, a Skill that uploads the audio from the model's sandbox without the audio passing through the LLM. See Connectors.

And an MCP server? Yes (audio by public URL only). For local files, use the Skill. See Connectors.

Slack and Telegram? Yes, there are bots that transcribe the audio you send them. See Connectors.

n8n? Yes, there are Uttera nodes for n8n.

Do you have apps to use it without coding? Yes, installable web apps: Summary, Interpreter (voice↔voice) and Pronunciation coach.

The Skill gives me a network error in Claude. On Team/Enterprise accounts an admin must allow api.uttera.ai in the organization's network allowlist. On individual plans it isn't needed. See Connectors.

Async and webhooks

Can I send long jobs without waiting? Yes: async mode queues the job and notifies you by webhook when done, without the sync 10-minute cap.

Are webhooks signed? Yes, with HMAC, so you can verify they come from us.