Raptor is a governance layer for AI systems built for DoD and DISA environments. Every response carries trust boundaries, passes confirmation gates, and is traceable to its source — classified, hash-chained, and auditable by design.
4
AI Providers
11
Immutable Tables
32
API Routes
Every request passes through five deterministic stages. The AI generates content. Everything else — classification, enforcement, trust marking, and hash-chaining — is rules-based and auditable.
Deterministic pattern matching assigns intent: respond, ask, propose action, or execute confirmed action.
Middleware checks capabilities, policies, budgets, rate limits, and suspension status.
AI generates content in a sandboxed proposal layer. It cannot commit or execute.
Deterministic layer validates, assigns trust boundaries, hashes, and chains to prior event.
Committed, classified, hash-chained response reaches the user. Fully replayable.
Every segment of every response is tagged with its epistemic status. The weakest boundary in a response dominates the response-level classification. No ambiguity about what's verified and what isn't.
Verified against authoritative source
Action ran with recorded result
Fetched from system of record
High-confidence inference, not verified
Limited evidence available
Observability logs after the fact. Evals test capability. Raptor governs what's permitted to happen — deterministically, before it reaches the user.
SHA-256 hash chain across 11 append-only Postgres tables. Database-level triggers prevent UPDATE and DELETE — not application-level protection. Standalone verification without system access.
Hard architectural separation: AI proposes, deterministic core commits. Prevents prompt injection reaching execution, hallucination reaching operations, and unaudited side effects.
Actions require explicit human confirmation before execution. The governance layer does not use AI to make governance decisions. AI generates content — everything else is rules-based.
Anthropic, OpenAI, Gemini, Together AI. Switching providers does not change governance, trust boundaries, or audit chain. Health monitoring with circuit breaker and auto-failover.
Cryptographically signed, versioned, append-only governance policies. Enforced or permissive modes. Visual policy editor with form-based rule builder and one-click sign + activate.
PostgreSQL RLS on 30 tables with FORCE ROW LEVEL SECURITY — database-level, not application-level. Three roles: Owner, Admin, Member. Identity via Clerk SSO.
Built for the regulatory frameworks that govern AI in DoD and DISA environments.
Pay for governed responses. Provider cost pass-through plus a governance fee. No seats, no tiers, no surprises.
Plus provider cost + 10%. $10 minimum credit purchase. No expiration.
500 governed responses, 1 seat. Full governance lifecycle. No credit card required.
Start a 14-day trial or send us your questions, comments, and concerns.
