Built around the context boundary.

Multra.ai uses tenant-scoped access, hashed API keys, a tenant-scoped provider secret store, audit logs, and context firewall policies.

Provider credentials

Credentials are stored only in the Context Kernel secret store for the owning tenant. Plaintext keys are never returned after creation or copied into the application database.

Request privacy

• Provider-call content is processed transiently.
• Eligible compiled checkpoints may use a bounded, tenant-scoped process-local cache; privacy and retention policy can disable it.
• Receipts and analytics retain bounded metrics, states, cryptographic identities, and pseudonymous workflow identifiers, not raw prompts, checkpoints, source code, or model output.
• Repository graphs and temporary workbench files are stored only when those features are explicitly used.

Important honesty

Context Kernel and the selected provider must process request content in plaintext to answer it. Pseudonymization hides selected identifiers, but semantic meaning can still reveal sensitive information.