CloudLast verified June 29, 2026

Cohere data retention policy

Cohere may use API prompts and generations to train its models by default, with a dashboard opt-out, and retains logged data for 30 days on the SaaS Platform. Last checked 2026-06-29 against Cohere's published commitments, plus how to route Cohere through Meetily as a BYOK summary provider via its OpenAI-compatible endpoint.

Quick policy snapshot

Default retention: 30 days (default)
Zero data retention available: Yes
Trains on API customer data by default: Yes

Cohere's API training default

Cohere's default posture is different from most other commercial API providers documented on this site. Under Cohere's Enterprise Data Commitments, prompts and generations on the SaaS Platform may be used to train Cohere models unless you opt out. Providers such as Anthropic and OpenAI take the opposite default on their paid APIs, where customer content is not used for training without an explicit opt-in.

Opting out is straightforward and available at any time. In the Cohere Platform, open Settings, select Data Controls in the left menu, and set the training toggle to Off. Cohere also states that it filters and strips common types of personal information from prompts and generations before any are made available for possible training use, but the cleaner control for a privacy-sensitive workload is to opt out entirely.

Because this is an on-by-default setting, it is a step the account owner must take. Meetily routes traffic under your own API key and cannot change your Cohere data controls for you.

Retention

On the SaaS Platform, Cohere automatically deletes logged prompts and generations after 30 days. The Enterprise Data Commitments name two exceptions: a longer period where a legal requirement or a customer contract requires it, or where usage is flagged as potentially violating Cohere's terms, including its Usage Policy.

These commitments apply to commercial, paying customers. For the SaaS Platform, that means accounts with a credit card on file. Trial usage falls under Cohere's standard Terms of Use and Privacy Policy instead, so confirm which set of terms governs your account before relying on the 30-day window.

Zero data retention

Cohere offers zero data retention for eligible enterprise customers. Under ZDR, Cohere does not log any prompts or generations at all. Because ZDR removes Cohere's normal ability to monitor for misuse and abuse, it is offered only to enterprise customers who can make additional commitments about their usage. To request it, contact support@cohere.com and copy your sales representative if you have one. ZDR is not a self-serve dashboard toggle, and Cohere notes that usage data is still received even when ZDR is enabled.

How Meetily uses Cohere

Meetily transcription is always 100% local, and audio never leaves your device. For summaries, Meetily's BYOK flow accepts any OpenAI-compatible base URL.

Cohere's native API at api.cohere.com uses Cohere's own proprietary request shape and is not OpenAI-compatible. To pair Cohere with Meetily, use Cohere's separate OpenAI-compatible endpoint at https://api.cohere.ai/compatibility/v1, which Cohere publishes specifically so applications built for the OpenAI SDK can call Cohere models. You set that compatibility base URL in Meetily's BYOK configuration and supply your own Cohere API key. Meetily then sends transcript text, not audio, over TLS for summarization, and the response is stored locally on your device.

Your account's contract terms and the defaults on this page apply to that traffic. Given that Cohere trains on API data by default, opt out in Data Controls before sending real meeting content, or switch your Meetily summary model to a local provider like Ollama to keep the entire pipeline on-device.

References

"Enterprise Data Commitments", Cohere, https://cohere.com/enterprise-data-commitments (accessed 2026-06-29). Primary source for the 30-day SaaS retention window, the training-by-default posture and dashboard opt-out, ZDR eligibility for enterprise customers, and the commercial paying-customer scope.
"AI Security and Data Protection", Cohere, https://cohere.com/security (accessed 2026-06-29). Cross-cloud deployment posture and data-control framing; no fixed public processing region.
"Cohere Privacy Policy", Cohere, https://cohere.com/privacy (accessed 2026-06-29). Multi-jurisdictional operations (Canada, EU, US) and Data Processing Addendum reference.
"Using Cohere models via the OpenAI SDK", Cohere Documentation, https://docs.cohere.com/docs/compatibility-api (accessed 2026-06-29). Confirms the OpenAI-compatible base URL https://api.cohere.ai/compatibility/v1 used with a Cohere API key, distinct from the native API.

Last verified: June 29, 2026. Policy source: Cohere policy

Frequently asked questions

Does Cohere train its models on my API prompts and outputs?

By default, yes. Cohere's published commitments state that prompts and generations may be used to train Cohere models unless you opt out. This is the opposite default from providers like Anthropic and OpenAI, whose paid APIs do not train on customer content by default. You can disable training in the Cohere dashboard under Settings, Data Controls, by switching the toggle to Off. Confirm the current default against the linked policy before relying on it for a regulated workload.

How do I opt out of Cohere using my data for training?

Open the Cohere Platform, go to Settings, then Data Controls in the left menu, and set the training toggle to Off. Cohere documents this as available at any time. Because training is on by default, this is a step you must take yourself; Meetily cannot toggle it for you.

How long does Cohere retain my API prompts and outputs?

Cohere automatically deletes logged prompts and generations after 30 days on the SaaS Platform, unless a legal requirement or a customer contract requires longer retention, or unless usage is flagged as potentially violating their Usage Policy. The 30-day window and its exceptions are stated in Cohere's Enterprise Data Commitments.

Is zero data retention available for the Cohere API?

Yes, but only for eligible enterprise customers. Under zero data retention (ZDR), Cohere does not log any prompts or generations. Because ZDR removes Cohere's ability to monitor for misuse in the usual way, it is offered only to enterprise customers who make additional commitments about their usage. Request it by contacting support@cohere.com, copying your sales representative if you have one. ZDR is not a self-serve toggle.

Does the training opt-out cover only paid API traffic?

Cohere's Enterprise Data Commitments, including the 30-day retention window, apply to commercial, paying customers, which for the SaaS Platform means accounts with a credit card on file. Trial usage is governed by Cohere's standard Terms of Use and Privacy Policy instead, which can differ. Scope your expectations to whichever account type you are billing under.

Where is Cohere API data stored geographically?

Cohere operates across multiple jurisdictions including Canada, the EU, and the US, and publishes a multi-jurisdictional Data Processing Addendum. Cohere does not publish a single fixed processing region on its public pages, and its models can be deployed across several cloud platforms. For a specific residency commitment, check the current DPA and your account configuration with Cohere.

How does Meetily route to Cohere when I pick it as my summary provider?

Meetily transcription is always 100% local. For summaries, Meetily's BYOK flow accepts any OpenAI-compatible base URL. Cohere's native API at api.cohere.com is its own proprietary shape, but Cohere also publishes a separate OpenAI-compatible endpoint at https://api.cohere.ai/compatibility/v1. You point Meetily at that compatibility base URL with your own Cohere API key, and Meetily sends transcript text (not audio) for summarization. The retention and training defaults on this page apply to your traffic, not Meetily's.

What if I want zero retention but my account does not have ZDR?

Two options. Ask Cohere to enable zero data retention on your enterprise account, and turn off the training toggle in Data Controls in the meantime. Or switch your Meetily summary model to a local provider like Ollama, which keeps the entire pipeline on your device. Meetily's transcription path is already local, so the local-summary route never sends transcript text off-device.

Use Cohere with Meetily, on your terms

Meetily transcription stays 100% local. For summaries, bring your own Cohere key (BYOK) so the data path matches the policy you just read - or pick a local model if you want zero retention by construction.

Set up BYOK Download Meetily