terms of service.

A streaming speech-to-text API for voice agents. You open a websocket, send audio, and receive interim and final transcripts as you speak. The endpoint is compatible with the Deepgram streaming protocol, so most existing clients work by changing the URL and key.

The service is in private beta. Interfaces and pricing may change, and we do not offer an uptime guarantee unless we have signed one with you separately.

• Your API key is shown once when you create it. Keep it secret — anyone holding it can spend your credit.
• Never embed a long-lived key in a browser or mobile app. Use the short-lived tokens the console issues instead.
• You are responsible for all usage on your keys until you revoke them, which takes effect immediately.
• One account per individual or organisation. Don't resell access as if it were your own service.

Usage is metered by the second of audio processed. New accounts start with a small free credit; after that you add credit by card through Stripe, and calls draw against it.

• Prepaid accounts: when credit runs out, in-flight calls are ended and new calls are refused. Watch your balance.
• Invoiced (enterprise) accounts: calls are never interrupted for credit reasons, and we bill from recorded usage.
• Rejected and failed calls are recorded but not charged for audio they never processed.
• Credits are for use with the service, do not expire while your account is active, and are not redeemable for cash.

Your usage dashboard is the record of what you have consumed. If you think it is wrong, tell us and we will look into it.

Transcription is easy to point at people who have not agreed to it. You must have a lawful basis — and, where your jurisdiction requires it, notice or consent — to record and transcribe every person whose speech you send us. That obligation is yours, not ours.

Don't use Autoloops to:

• Record or transcribe people unlawfully, or run covert surveillance.
• Build profiles of individuals from their speech without their knowledge, or attempt biometric voice identification.
• Process audio you have no right to process.
• Harass, defraud or harm others, or violate any law or third-party right.
• Extract, copy or reverse-engineer the underlying models, or scrape the service at scale.
• Deliberately overload the service or evade rate limits and credit checks.

You own the audio you send and the transcripts we return. We claim no rights over them.

We do not store audio or transcripts, and we do not train on them — see the privacy policy for exactly what we keep.

Speech recognition is probabilistic. Transcripts will contain errors, especially with background noise, overlapping speakers, accents and unusual vocabulary. Don't rely on them unreviewed for medical, legal, financial or safety-critical decisions.

We aim for high uptime but do not guarantee uninterrupted service. We may perform maintenance, change or withdraw features, and apply rate limits. We will give reasonable notice of material changes, and at least 14 days' notice of price changes.

To the maximum extent permitted by law, the service is provided "as is" without warranties. We are not liable for indirect, incidental or consequential damages, or for decisions made on the basis of a transcript. Our total liability is limited to the amount you paid us in the three months before the claim.

You can stop using Autoloops at any time; unused credit on a closed account is not refunded unless we agreed otherwise in writing. We may suspend accounts that breach these terms or that put the service at risk, and will tell you why where we can.

Questions: legal@autoloops.ai.