Private inference

Private inference is the foundation for building with Mystro. On supported routes, TEE execution and E2EE protect the inference path. Confirm the exact model route and its supported protections during invite-led onboarding.

Zero-retention model processing

Mystro private inference routes use zero-retention model processing for inference-provider prompts and responses, including fallbacks. Application history, logs, stored knowledge, and external tool data have separate retention policies.

Choose the boundary before the model

Start with the data your workflow needs and who may see it. Then confirm that an approved route satisfies those requirements. A model name alone does not establish a protected execution path.

A private inference route does not make the rest of a workflow private automatically. Your application receives results, tools may receive selected arguments, and stored records have their own access and retention boundaries.

Build with explicit decisions

• Confirm the selected route's TEE and E2EE coverage and verification details.
• Minimize prompts and tool arguments to the data needed for the task.
• Agree on behavior when that route is unavailable. Do not silently fall back to a route with weaker protections.
• Review storage, logging, key ownership, and retention separately.

Continue with TEE and E2EE, data boundaries, and model routes.