Culvii was designed for environments where a security failure isn't a bad week — it's a regulatory event. Here's exactly how we think about security, what we've built, and what we're still working toward.
Every agent is an Ed25519 keypair. Every action is signed. Trust is established cryptographically, not through network position, IP allow-lists, or database checks. Compromising a single agent does not compromise your tenant, other tenants, or the registry.
An agent can only execute within its explicitly granted capability set. Capability grants are time-bounded and revocable. There is no way for an agent to escalate its own privileges — trust elevation requires an out-of-band human action.
Every observation event is chained and anchored to a Merkle Tree. You can verify the integrity of any agent's audit log at any time. The audit log cannot be modified retroactively without detection. Same cryptographic primitive as certificate transparency logs.
We believe in honesty about compliance. Here is the accurate current status of each standard — not aspirational marketing.
| Standard | Status | Notes |
|---|---|---|
| SOC 2 Type II | In progress | Audit underway. Report will be available under NDA to enterprise customers upon completion. |
| GDPR | Compliant | Data Processing Agreements (DPAs) available. EU data residency supported in cloud deployments. |
| HIPAA | BAAs available | Business Associate Agreements available for Enterprise customers. Architecture supports HIPAA-compliant deployments with appropriate configuration. |
| ISO 27001 | Roadmap | Planned. Timeline to be confirmed. |
| PCI DSS | N/A | Culvii does not process, store, or transmit payment card data. |
All stored data — agent records, audit logs, cryptographic keys — encrypted at rest. No plaintext storage of sensitive material.
All connections require TLS 1.3. No unencrypted connections accepted. TLS 1.2 and below are explicitly blocked.
Agent private keys stored in HSMs in Culvii Cloud deployments. Isolated per tenant. Customer-managed KMS (AWS KMS, Azure Key Vault, HashiCorp Vault) available for Enterprise customers.
Stored separately from the main application database. Append-only at the storage layer. Cryptographically chained. No administrative path exists to modify historical entries.
All services run in private VPCs. No direct public exposure except load balancers and API endpoints. Internal service communication uses mutual TLS (mTLS).
Automated dependency scanning on every commit. Container image scanning in the CI/CD pipeline. Critical CVEs addressed within 24 hours.
Annual third-party penetration tests by an independent security firm. Summary report available to Enterprise customers under NDA.
Zero standing access to production systems. All access requires MFA, is time-bounded, and is fully logged. Engineering access follows strict least-privilege principles.
Every tenant is a hard boundary enforced at the database, application, and cryptographic layers simultaneously.
Separate schemas or separate databases per tenant depending on deployment tier. No shared tables or cross-tenant queries possible.
Each tenant has its own key namespace. No cross-tenant key sharing, ever. Tenant keys cannot be used to decrypt or sign another tenant's data.
Completely separate per tenant. A tenant's audit log is inaccessible to any other tenant, including Culvii employees, except under explicit legal process.
Capability namespaces are tenant-scoped. Cross-tenant capability grants require explicit configuration by an authorised tenant administrator.
Include: description of the vulnerability, steps to reproduce, and potential impact. PGP key available on request for sensitive disclosures.
| Severity | Definition | Customer notification |
|---|---|---|
| P0 · Critical | Service unavailable or confirmed data breach affecting customer data | <30 minutes |
| P1 · High | Significantly degraded service affecting production workloads | <2 hours |
| P2 · Medium | Non-critical functionality impaired; workaround available | <24 hours |
| P3 · Low | Minor issues with no material customer impact | Next business day |
Post-incident reports published on our status page within 5 business days of P0 and P1 incidents.
Send us your InfoSec questionnaire and we'll return it within 5 business days — fully completed, not just acknowledged. We also have a pre-filled SIG Lite for faster turnaround.