The security shape, from the operator’s seat.

The full Confidentiality and Security posture is on the Trust & Security foundation page. This page is the short version that an operator needs while running the platform. It does not replace Trust. It indexes into it.

The Cloud Harness installs the platform inside your own Cloudflare, AWS, Azure, or on-premise environment. The two stores, the enrichment workers, the MCP server, and the operator console all live in your account. Data does not leave your perimeter to be processed. If you end the vendor relationship with us, the install keeps running. Detail and contract language are on the Trust hub.

Security is enforced by the data access path itself. There is no separate policy layer between the request and the data. Every session ticket carries a scope; that scope cascades into which MCP tools resolve, which graph nodes can be traversed, which Postgres rows can be read, and which verbatim chunks resolve when cited. An unauthorized request returns no signal that anything was filtered. The mechanism is described in full under the Security Cascade.

Voice runs no-retention on every plan and every tier through our contract with ElevenLabs. The LLM layer runs no-training on every plan and no-retention on Enterprise, through our contract with Anthropic. Both properties are in signed contracts before any data flows. The audio handling chain (zero-retention contract, in-memory worker, synthetic voice substitution) is described under Synthetic Voice Substitution.

In accordance with Anthropic’s usage policy and compliance, we do not work on military applications. We do not support surveillance of named individuals; any OAuth scope that would enable it is rejected before any data flows. We do not help anyone re-identify a respondent; the voice substitution chain and the security cascade leave no handle to attempt it from. The full list, and the architectural enforcement point behind each refusal, is under the Refusal Protocol.

Encryption at rest and in transit, tenant-scoped secrets, per-environment KMS keys, the documented incident response procedure, the responsible disclosure program, the subprocessor list, and the DPA are all covered on the Trust hub. The fail-closed checklist applied at every boundary (webhook HMAC verification, per-tenant ownership, token budgets, schema validation) and the prompt-integrity hash chain are in the engineering reference, available on request.