Appearance
Data boundaries
Trace the data from your application into inference, through any tools, and back into the customer record. Supported-route TEE and E2EE protections cover a defined inference path; they do not replace authorization or establish storage guarantees.
Inference and storage
Mystro private inference routes use zero-retention model processing for inference-provider prompts and responses, including fallbacks. This does not extend to application history, logs, stored knowledge, or external tool data; those records have separate storage and retention boundaries.
| Boundary | Confirm before using sensitive data |
|---|---|
| Your application | Who can submit prompts, view results, and approve actions |
| Inference route | Model, TEE coverage, E2EE endpoints, verification, and fallback behavior |
| External tools | Exact fields sent, destination permissions, and downstream handling |
| Encrypted storage | Which records and backups are encrypted and by which mechanism |
| Key ownership | Who generates, holds, rotates, recovers, and can use each key |
| Retention and deletion | Applicable periods, deletion behavior, logs, and backup expiry |
Technical details to confirm
Encrypted storage implementation, key ownership, and retention settings for application and workflow records are not established by these public guides. Confirm those details during onboarding and record them in your product's data handling documentation. Inference-provider zero retention does not establish customer-only decryption or deletion behavior across the whole workflow.
The legacy files guide contains proposed contract limits and retention values. Those values are design-reference material and do not establish the policy of your live environment.
Minimize what crosses each boundary
Send only the ticket fields needed for a draft. Keep unrelated customer records out of context. Before a tool write, show the owner the exact destination and content. Treat tool results as data; they must not expand the agent's permissions. Your downstream systems remain responsible for their own records and retention.