Your Data at Deepgram

Your data is yours. This page explains what Deepgram does with the audio, text, and transcripts you send us, and the controls you have over each part of it.

Three things determine how your data is handled:

• Where it is processed. The endpoint you call sets the geography. The global endpoint is the default. Regional endpoints handle your request and store your data inside one region. Model improvement on retained data can happen outside that region, so processing is fully in-region only when the request is also opted out.
• Whether it is retained and used to improve our models. Requests participate in the Model Improvement Program (MIP) by default, and Deepgram retains them. Retained data may be processed outside your region for model improvement. Opt out per request with mip_opt_out=true. Opted-out requests are not retained, and on a regional endpoint Deepgram does not send them outside the region (third-party Voice Agent providers are separate).
• How it is protected. Your data is encrypted in transit and at rest, and is processed only by Deepgram and our published sub-processors, or by providers you choose using your own keys.

The two settings are separate. The residency guarantee depends on both.

Request metadata and usage logs record how you used the API. They do not contain your audio, text, or transcripts. To retain usage data past 90 days, fetch it on an interval with the Usage API and store it yourself. See Using Logs & Usage.

Deepgram does not sell your data, does not use it to build advertising profiles, and does not redistribute it to third parties without your permission.

The Model Improvement Program (MIP) is how Deepgram improves its models on real-world audio. Participation is the default, and participants receive models tuned to their domain, faster support, and early access to new models.

Data used for training is only the data included through MIP. It is never sold, never used for advertising, and never redistributed to third parties without your permission.

Set mip_opt_out=true and the request is excluded from model improvement:

For example, to opt a pre-recorded transcription request out of model improvement:

curl -X POST "https://api.deepgram.com/v1/listen?model=nova-3&mip_opt_out=true" \
  -H "Authorization: Token YOUR_DEEPGRAM_API_KEY" \
  -H "Content-Type: audio/wav" \
  --data-binary @audio.wav

Opting out gives you zero data retention: Deepgram does not store your audio, text, transcripts, or synthesized audio after the response is returned. Request metadata and usage logs are still recorded, as described above.

For per-language and per-SDK examples, see Model Improvement Program.

Check the Deepgram Console under Usage > Logs, where an opted-out request shows mip_opt_out as true. You can also read it programmatically with the Get a Project Request endpoint.

If you would rather not set the flag on every request, contact us about enforcing opt-out across your account or deployment.

Where Deepgram processes your data depends on the endpoint you call.

api.deepgram.com is Deepgram’s global endpoint and the default for every API. It carries the broadest model and feature coverage, and it is the right choice for most applications. The Voice Agent API’s global host is agent.deepgram.com.

Data residency guarantees come from regional endpoints. If your data must stay in a specific geography, call that region’s endpoint.

Regional endpoints take the form api.{region}.deepgram.com. On a regional endpoint, Voice Agent traffic uses that region’s host: wss://api.{region}.deepgram.com/v1/agent/converse.

Each regional endpoint handles your request and stores your data within its region, and does not route requests to Deepgram services outside it. If the region is unavailable, requests fail rather than fall back to another region. For what model improvement adds, see the summary at the top of this page.

Regional endpoints use your existing API keys and SDKs. You only change the base URL. For endpoint URLs, WebSocket paths, and SDK configuration, see Regional Endpoints.

All regional endpoints support the same APIs:

• Speech-to-Text — Nova and Flux models (Whisper is not supported)
• Text-to-Speech — Aura and Flux models
• Voice Agent
• Text Intelligence

If you need processing pinned to a specific country rather than a region, use Deepgram Dedicated or self-hosted.

A regional endpoint keeps your data stored and processed in-region and stops routine offshore administrative access. It does not make Deepgram immune to foreign legal process. If your requirements go that far, deploy self-hosted or Deepgram Dedicated, where keys, logs, and retention stay in your own infrastructure.

Deepgram runs the listen and speak steps on Deepgram infrastructure in the region you call. If you configure a third-party LLM for the think step, that provider processes your data on its own infrastructure under its own residency and retention terms. Confirm those independently.

Where a managed provider offers a regional endpoint, Deepgram routes to it automatically. See Voice Agent managed providers for what is regionalized today.

Deepgram encrypts all data in transit and at rest with industry-standard encryption, including TLS 1.3 and AES-256.

Deepgram operates under SOC 2 Type 1 and Type 2 attestation, and supports HIPAA, PCI DSS, GDPR, CCPA, and the Australian Privacy Principles. For framework detail, regional specifics, and how to request documentation, see Data Privacy Compliance.

Deepgram publishes its sub-processors at deepgram.com/privacy/subprocessors.

If you have a residency or retention requirement this page does not cover, contact us.