Who CRP Is For¶

Context Relay Protocol is built for teams that ship AI in production and need it to be coherent, governed, and compliant - without rebuilding infrastructure from scratch.

Developers & Engineering Teams¶

The problem: You want to add context management and safety governance to your AI features, but building it yourself takes quarters.

CRP gives you:

One-line integration: Change base_url to https://gateway.crprotocol.io/v1
Progressive SDK: Start with zero concepts, unlock depth, tools, and knowledge control when you need them
Local-first: Runs on Ollama / LM Studio at $0 marginal cost
Multi-provider: OpenAI, Anthropic, Gemini, Bedrock, local - same interface

What changes in 30 days¶

Week 1: Drop-in governance via base_url or crp.SDKClient() - no app rewrite.
Week 2: Long-running tasks finish instead of truncating; quality scores replace guesswork.
Week 3: Safety Policy blocks HIGH/CRITICAL issues in CI before they reach production.
Week 4: Audit trail is queryable; compliance evidence is already generated from real calls.

import crp

client = crp.SDKClient()
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

Enterprises & Regulated Companies¶

The problem: EU AI Act fines are up to €35M or 7% of global turnover. You need audit-ready evidence, not consultant PDFs.

CRP gives you:

AIUC-1 aligned - 80%+ of AIUC-1 requirements covered out of the box, with a roadmap to accredited certification
EU AI Act Art. 6–17 coverage - risk classification, technical documentation, record-keeping, transparency, human oversight, accuracy
ISO 42001:2023 alignment - A.6.2.3–A.6.2.8 controls
Tamper-evident audit chain - HMAC-SHA256, per-window, never updated in place
Evidence packs - one-click regulator-ready bundles
SSO / SAML, data residency, private cloud

What changes in 30 days¶

Week 1: EU AI Act risk classifier labels every deployed AI system.
Week 2: DPIA, technical docs, and transparency declarations are generated from runtime evidence.
Week 3: Auditor receives a tamper-evident evidence pack instead of static PDFs.
Week 4: Data residency and SSO policies are enforced across all governed calls.

CRP Comply

CISOs & Compliance Officers¶

The problem: You can't prove your AI system is safe because there's no runtime evidence.

CRP gives you:

Real-time risk scoring on every call - LOW / MEDIUM / HIGH / CRITICAL
Control evidence - HMAC-signed proof of every safety and governance decision for EU AI Act, AIUC-1, ISO 42001, NIST, and SOC 2-for-AI
AIUC-1 mapped controls - adversarial robustness, filtering, harmful-output prevention, and audit
Hallucination detection - 13-stage DPE pipeline
PII scanning - 7 categories, hashed never stored raw
Safety Policy enforcement - configurable halt levels
Compliance dashboard - live view of every governed call

What changes in 30 days¶

Week 1: Every LLM call returns a risk score and grounding signal.
Week 2: HIGH/CRITICAL outputs are halted or checkpointed automatically.
Week 3: PII and prompt-injection attempts are flagged before leaving the Gateway.
Week 4: Incident response time drops from days to minutes with session reconstruction.

The Safety Case

Startups & Indie Builders¶

The problem: You need AI governance but can't afford enterprise tools or dedicated compliance staff.

CRP gives you:

Free tier: 100 governed calls/mo, no card required
Starter: $49/mo ($490/yr) - 5,000 calls, checkpoint inbox, evidence packs
No seat-based pricing - pay for usage, not headcount
One-person setup - configure safety policy via no-code console in minutes

What changes in 30 days¶

Week 1: Safety, audit, and continuation work out of the box with one integration line.
Week 2: Long-form outputs and agent chains become shippable instead of experimental.
Week 3: Compliance evidence is ready for investor and customer security reviews.
Week 4: Usage scales to Team or Scale tiers without re-architecting governance.

Pricing

AI Product Managers¶

The problem: You are shipping governed AI features slowly because safety, compliance, and quality reviews happen late in the cycle.

CRP gives you:

Faster launches - safety and provenance are built in from the first call, not bolted on before release
Built-in safety - 13-stage DPE scoring, safety profiles, and human checkpoints reduce late-stage risk surprises
Compliance evidence - EU AI Act, ISO 42001, and GDPR artifacts generated from runtime data
Clear quality metrics - answer.quality, answer.complete, answer.crp.risk, and source attribution give stakeholders an objective signal

What changes in 30 days¶

Week 1: Define a safety policy that follows the feature into every environment.
Week 2: Replace "we think it works" with S/A/B/C/D quality tiers on every output.
Week 3: Compliance, legal, and security review cycles shrink from weeks to days.
Week 4: Ship the feature with regulator-ready evidence and a tamper-evident audit trail.

SDK Guide

AI Researchers & Data Scientists¶

The problem: You need to compare context strategies and reproduce results.

CRP gives you:

Built-in benchmarking - compare CRP vs RAG vs Injection vs Hierarchical
SQB harness - Semantic Quality Benchmark with per-window metrics
Reproducible pipelines - every run is HMAC-signed and auditable
Local models - test on Ollama, LM Studio, vLLM without API costs

Benchmarks

Open Source & Standards Contributors¶

The problem: You want an open, vendor-neutral protocol for AI context and safety.

CRP gives you:

Open specification - 50 specs, CC BY 4.0
Submitted to IETF, IANA, IEEE SA, ISO/IEC JTC 1/SC 42
Reference implementation - Elastic License 2.0
Conformance suite - test your implementation against the protocol

Read the Spec

Concrete use cases¶

Long-form report generation¶

Continuation engine + voice profile + document map + re-grounding.
Outcome: 30-section reports completed with a conclusion instead of truncating at section 8.

Penetration testing workflows¶

Each recon/tool-select/analysis/report step gets a fresh, fact-packed window.
Outcome: Tool outputs become persistent, ranked findings; final report carries all discoveries.

Legal / contract analysis¶

Auto-ingests million-token documents; CKF links clauses across pages.
Outcome: Cross-reference questions answered with source attribution.

Medical literature review¶

Cross-session CKF accumulates papers; community detection groups related findings.
Outcome: Synthesized evidence with provenance back to each paper.

Customer-support automation¶

PII detection, GDPR consent/retention records, EU AI Act risk classification.
Outcome: Regulated chatbot deployment with audit-ready evidence.

Code generation across large codebases¶

Multi-window generation with architecture facts in each envelope.
Outcome: Coherent multi-file output with structural integrity.

Small-model deployment¶

ORC/ICML/RTL scaffolding for 2B–7B local models.
Outcome: Local models perform reasoning tasks normally requiring 10× larger models.

CI/CD governance scanning¶

CRP Scan GitHub Action flags ungoverned LLM calls and opens remediation PRs.
Outcome: HIGH/CRITICAL AI governance gaps caught before merge.

See the full capabilities reference