CRP Gateway¶

One line of code turns any LLM call into a governed, audited, continuation-aware AI system.

CRP Gateway is a drop-in HTTPS proxy that sits between your application and any OpenAI-compatible provider. It adds the Decision Provenance Engine (DPE), Safety Control Plane, context envelope packing, automatic continuation, and tamper-evident audit trails - without touching your application logic.

Status

Managed cloud at gateway.crprotocol.io is on the waitlist. Self-hosted and white-label deployments are available for Enterprise. Contact sales →

The value in one minute¶

Without Gateway	With Gateway
Every LLM call is a black box	Every call returns risk, grounding, and provenance
Output truncates; task stalls	Continuation engine finishes multi-section work automatically
Safety code is optional and scattered	Safety policy is enforced at the network boundary
Compliance evidence is manual	HMAC-chained audit events stream to CRP Comply
Context is flat text	CKF-scored envelope puts the most relevant facts first

Why teams choose CRP Gateway¶

Non-bypassable safety. Because enforcement lives at the base_url boundary, application code cannot accidentally skip the DPE pipeline or Safety Policy evaluation.
Finish long tasks automatically. The continuation engine detects output limits and dispatches fresh windows until the task is complete - 11.8× more output on the same model.
One integration, every provider. OpenAI, Anthropic, Gemini, Ollama, vLLM, and any OpenAI-compatible endpoint get the same governance contract.

Who it's for¶

Role	Problem	How Gateway helps
Platform Engineer	Need governance without a full app rewrite	Change `base_url`; get safety, audit, and continuation instantly
AI Product Manager	Safety reviews slow down launches	Ship with risk scores and provenance from the first call
CISO / Security Lead	Can't enforce safety across scattered services	Centralised policy enforcement at the network boundary
Startup CTO	Need enterprise-grade governance on a budget	Start at $29/mo and scale usage without re-architecting

How it works¶

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

Your app sends a normal OpenAI request to the Gateway URL.
Gateway packs context using CDR/CDGR retrieval and the maximally-saturated envelope.
Safety Control Plane runs 13+ checks in <50ms.
DPE scores hallucination risk, grounding, fabrications, and contradictions.
If output hits a window limit, the continuation engine starts the next window automatically.
HMAC-signed audit events are emitted to your configured store or CRP Comply.
Your app receives a normal response with extra CRP-* headers containing governance metadata.

Integration options¶

Option A: Change `base_url` (zero code changes)¶

from openai import OpenAI

client = OpenAI(
    base_url="https://your-gateway.example/v1",  # or your self-hosted URL
    api_key="your-provider-api-key"
)

# Every call is now risk-scored, audited, and continuation-aware
response = client.chat.completions.create(model="gpt-4o", messages=[...])

Option B: Use the native Python SDK¶

import crp

client = crp.SDKClient(provider="openai", model="gpt-4o")
client.ingest("./docs/")
answer = client.ask("Write a deployment guide", depth="thorough")

print(answer.text)
print(answer.quality)   # S | A | B | C | D
print(answer.sources)   # cited facts from your documents
print(answer.crp.risk)  # LOW | MEDIUM | HIGH | CRITICAL

SDK Guide

What you get on every call¶

Capability	What it does	Spec
Context envelope	Packs the best facts into a dedicated task window	SPEC-003
CDR / CDGR retrieval	Scored, graph-walked retrieval from CKF	SPEC-005
Continuation engine	Auto-continues when output limits hit	SPEC-004
Safety policy enforcement	Configurable halt/flag/audit actions	SPEC-006
DPE scoring	Risk, grounding, fabrication, contradiction signals	SPEC-005
Audit trail	HMAC-chained, tamper-evident event log	SPEC-011
Multi-agent safety	Budget headers, chain provenance, cross-agent controls	SPEC-012

Deployment options¶

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

Self-hosting options →

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.

Get started¶

Use CRP Gateway self-hosted today, join the managed-cloud waitlist, or talk to sales about white-label deployments.

Join waitlist → View Pricing → SDK Guide →

Pricing¶

See the Pricing page for current Gateway plans and quotas.