CRP Gateway¶

Managed AI safety infrastructure. A drop-in HTTPS endpoint that adds the Decision Provenance Engine, Safety Policy enforcement, and audit-trail emission to any OpenAI-compatible API call — with a single base-URL change.

Status

Beta — selected enterprise customers. Open beta planned for 2026 Q3. Request access →

What It Does¶

Terminates inbound HTTPS at a CRP-conformant gateway.
Runs the full DPE pipeline (SPEC-005) on every response.
Enforces CRP-Safety-Policy directives (SPEC-006) at the transport layer.
Emits HMAC-chained audit records (SPEC-011) to your auditor of choice.
Streams compliance evidence into CRP Comply automatically.

Why a Gateway¶

Application-level safety code is uneven, optional, and easy to forget. A gateway makes safety non-bypassable for any service that points its LLM client at the gateway URL.

┌──────────────┐     base_url = gateway     ┌─────────────────┐     vendor SDK     ┌────────────┐
│ Your App     │ ─────────────────────────► │ CRP Gateway     │ ─────────────────► │ OpenAI /   │
│ (no changes) │ ◄───────────────────────── │ DPE + Policy +  │ ◄───────────────── │ Anthropic /│
└──────────────┘   CRP-* headers added      │ Audit + Comply  │                    │ Local /…   │
                                            └─────────────────┘

Deployment Options¶

Mode	When to use
Managed cloud (`gateway.crprotocol.io`)	Fastest path. Multi-tenant, single-tenant, or sovereign region.
Self-hosted container	You want the gateway inside your VPC / on-prem.
Sidecar	Per-pod injection in Kubernetes.
Edge	Cloudflare Worker / Fastly Compute deployment for global low-latency.

Compatible APIs¶

OpenAI Chat Completions, Anthropic Messages, Google Gemini, Mistral, Cohere, Ollama, LM Studio, vLLM, TGI, and any OpenAI-compatible endpoint.

Conformance¶

CRP Gateway is Level 2 conformant per SPEC-014.

Pricing¶

Pricing is per million governed tokens, with volume bands and a free development tier. Contact gateway@crprotocol.io for pricing.

Request access