Zero-access by design

Only you and the teammates you authorize. Tresor staff cannot access your content, and our cloud providers cannot access it either.

Tresor is designed so that your organization controls the keys needed to access content. In practice, keys are issued and shared with your authorized users (for example by wrapping keys per user), rather than being available to Tresor operators.

Not in the "messaging app" sense. In classic end-to-end encryption, the server never sees plaintext. In Tresor, plaintext is processed inside secure enclaves so the AI can work. The important part is that plaintext is not available to Tresor staff or the cloud provider, and data is encrypted outside enclaves.

Your workspace is designed to run with EU data residency, and sensitive processing happens in EU-hosted confidential computing environments.

Tresor is built for teams. Access is controlled by your organization: invite someone and they can be granted access to specific projects; remove someone and access can be revoked. You can apply role-based controls (for example who can create projects, invite members, export content).

To operate a service, some minimal data may exist outside the encrypted content layer, such as: account and billing data, seat count and plan, basic system events needed for reliability and abuse prevention. Tresor's goal is to keep user content and sensitive labels outside operator reach.

Zero-access is meant to remove that dependency. Support should not require content access. If you choose to share something (for example a snippet you paste into a ticket), that shared text is visible in that support channel.

Integrations can introduce new data paths. Tresor can keep its own platform zero-access, but if you connect external tools (storage, email, CRMs), those tools have their own access and retention rules. The safest approach is to treat connectors as explicit trust decisions.

Tresor can provide verifiable signals (for example attestation evidence or receipts) showing that requests were processed in an approved confidential runtime. This makes "by design" more than a promise.

Zero-access does not protect against: a compromised user device, an authorized teammate copying data elsewhere, or content you intentionally export or send to third parties. It's designed to prevent the most common failure mode in mainstream AI tools: provider and cloud-operator access to customer content.