🎯 The Agentic Positioning Layer for SLM-first AI

Position every agent on the right task,
with the right context and tools. ¶

MCP exposes tools. A2A connects agents. Context Relay Protocol™ positions every agent. One base_url change turns small local models into capable, governed agents: CRP selects operations, loads only the 1–3 tools each call needs, carries state forward, and emits HMAC-signed proof that safety, grounding, and policy controls ran. The evidence the EU AI Act, AIUC-1, ISO 42001, and NIST AI RMF all demand, generated from live runtime data.

Position your agents free in 60 seconds See the control-evidence mapping

13-stageSafety on every call

<50msOverhead

HMACSigned audit chain

EU AI Act + AIUC-1Control evidence

Controls are easy to claim. CRP proves they operate.¶

Every major AI assurance framework - EU AI Act, AIUC-1, ISO/IEC 42001, NIST AI RMF, SOC 2-for-AI - demands the same three things underneath its own vocabulary:

Controls exist - you have safety, security, and governance mechanisms.
Controls operate - they run on every AI call, continuously, not on paper.
You can prove it - verifiable evidence the controls ran, not assertions.

Most organisations manage #1 and sometimes #2. The universal failure is #3. CRP closes that gap: every governed AI call emits signed, tamper-evident evidence that your security and safety controls ran - the proof all of these standards require.

Different standards. One agentic positioning engine.

See exactly which CRP capabilities produce evidence for which standard.

Control-evidence mapping AIUC-1 proof point

Outcomes & ROI¶

📝 Finish long tasks¶

11.8× more content completed without rewriting prompts.

Automatic continuation turns truncated 8-section outputs into 25-section deliverables that actually conclude.

⚖️ Cut compliance time¶

Evidence packs generated in seconds, not months.

EU AI Act, ISO 42001, GDPR, and NIST AI RMF reports are built from runtime audit data, not consultant interviews.

🛡️ Reduce hallucination risk¶

13 DPE signals on every call in <50 ms.

Fabrications, contradictions, grounding failures, and prompt injections are caught before they reach users.

🚀 Deploy faster¶

One line of code to governed AI.

Change base_url or call crp.SDKClient(); no new infrastructure, no provider lock-in.

Trusted by teams building with AI¶

AI-native startups Regulated enterprises Consulting firms Security teams

Customer case studies and logos are coming soon. Share your story.

Try it in 5 lines¶

5 lines to governed AI

import crp

client = crp.SDKClient()
client.ingest("./docs/")
answer = client.ask("Write a complete 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

That is the full pipeline: ingestion → CDR/CDGR retrieval → envelope packing → safety scan → LLM dispatch → continuation → extraction → provenance → source attribution. No hidden heuristics. Every score is coupled to an enforceable action.

Run the interactive demo locally →

What CRP solves¶

Context dies at 8 sections¶

Your model stops mid-task. Previous context is lost. There is no continuation.
CRP automatically continues across windows until the task is done, then stitches outputs into one coherent result.

No safety guardrails¶

Every AI call returns raw output. No risk score. No provenance. No audit trail.
CRP runs a 13-stage safety check on every call in under 50ms and records every event to a tamper-evident chain.

Compliance is manual¶

EU AI Act, ISO 42001, GDPR. You hire consultants for months.
CRP generates evidence packs automatically from runtime data - risk assessments, DPIAs, technical docs, transparency declarations.

Integration is expensive¶

Building context management, safety, and compliance from scratch takes quarters.
CRP is one line: base_url="https://your-gateway.example/v1" or crp.SDKClient().

Explore by topic¶

🛡️ AI Safety¶

Hallucination detection, prompt injection shield, PII scanning, and safety budgets on every call.

AI Safety

⚖️ AI Governance¶

Declarative policies, identity headers, human-in-the-loop checkpoints, and tamper-evident audit.

AI Governance

📋 AI Compliance¶

Automated EU AI Act, ISO 42001, NIST AI RMF, and GDPR evidence from runtime data.

AI Compliance

🧠 Context Management¶

Unbounded context, automatic continuation, graph-structured CKF, and retrieval for LLMs.

Context Management

🏅 Control Evidence¶

See which CRP capabilities produce signed, tamper-evident evidence for EU AI Act, AIUC-1, ISO 42001, and NIST AI RMF.

Control-evidence mapping

🚀 CRPv5 Roadmap¶

What's next: agentic positioning, SLM-first execution, closing certification gaps, and new SDK languages.

View Roadmap

Compare CRP with RAG, MCP & agents

Capabilities¶

CRP manages the full context lifecycle - ingestion, extraction, retrieval, packing, dispatch, continuation, safety, provenance, and compliance - in a single protocol layer.

Unbounded context¶

Automatic continuation, CSO relay, and re-grounding let tasks exceed any model's output limit without losing coherence.

Explore capabilities

6-stage extraction¶

Regex → NLP → NER → relations → discourse → LLM-assisted relational. Structured knowledge, not flat chunks.

Extraction protocol

13-stage DPE safety¶

Fabrication, distortion, contradiction, grounding, and hallucination-risk scoring in under 50 ms on every call.

DPE pipeline

Knowledge fabric¶

Graph-structured CKF with CDR/CDGR retrieval, multi-horizon context, and ephemeral scratch buffers.

CKF protocol

Two-sided provenance¶

Trace every claim back to its source and every input fact to its upstream origin, signed and tamper-evident.

Provenance

Compliance evidence¶

EU AI Act, ISO 42001, GDPR, NIST AI RMF evidence packs generated from runtime audit data.

Compliance

One-line integration¶

Drop-in Gateway base_url or native crp.SDKClient() with progressive SDK levels.

SDK guide

Agent-safe¶

Chain budgets, risk accumulators, circuit breakers, and clean MCP/A2A layering.

Multi-agent safety

Positioned tool loop (v5)¶

Turns a small local model into a capable agent by positioning, not injection: one operation at a time, only the 1–3 tools it needs, with a bounded per-operation window and typed state carried forward.

Explore capabilities

Read the full capabilities breakdown

The business case¶

11.8×More output

<50msSafety overhead

€35MPotential EU AI Act fine avoided

1 lineIntegration

Cost of unmanaged AI¶

Context truncation wastes expensive LLM tokens on repeated prompts.
Hallucinations damage customer trust and create liability.
Compliance reviews delay launches by weeks or months.
Security incidents require forensic log archaeology.

Value of CRP governance¶

Finish tasks in fewer total tokens despite using multiple windows.
Detect fabrications, injections, and PII before they reach users.
Ship with EU AI Act / ISO 42001 evidence already generated.
Reconstruct any answer from tamper-evident audit data in seconds.

Three products. One agentic positioning engine.¶

🌐 CRP Gateway¶

The runtime. Routes every AI call through governance + context management.

Managed-cloud waitlist · Self-host & white-label

Explore Gateway

📋 CRP Comply¶

The compliance layer. Generates audit-ready deliverables from Gateway evidence.

Managed-cloud waitlist · Self-host & white-label

Explore Comply

🔍 CRP Scan¶

The scanner. Finds ungoverned AI calls in your codebase and opens remediation PRs.

GitHub Action available · Managed cloud on the roadmap

Explore Scan

What's actually shipped¶

CRP is not a roadmap deck. The open-source SDK, CLI, and Scan action are live, documented, and shipping today:

🐍 Python SDK¶

4 progressive levels from one-line governance to full infrastructure control.

Drop-in base_url proxy
crp.SDKClient() with .complete(), .ask(), .ingest()
Depth negotiation: quick · standard · thorough · exhaustive
@client.tool automatic tool invocation + CKF extraction
Safety profiles: balanced · strict · permissive · research
client.audit.* and client.compliance.* evidence helpers

SDK Guide

🖥️ CRP CLI¶

Command-line governance tools for scanning, validation, and benchmarking.

python -m crp scan -- find ungoverned AI calls
python -m crp serve -- HTTP sidecar
python -m crp status -- session status
python -m crp preview -- envelope preview

CLI Guide

⚙️ GitHub Action¶

Auto-scan every PR for ungoverned AI calls with SARIF output.

Marketplace-published composite action
SARIF upload to Security tab
Auto-remediation PRs *(planned)*
VS Code extension *(planned)*

Scan Product

🌐 Gateway¶

OpenAI-compatible proxy with DPE, safety policy, and audit trails.

Managed-cloud waitlist; self-host and white-label available
58 CRP headers on every call
HMAC-chained audit events
Redis-backed session store

Gateway Product

📋 CRP Comply¶

Compliance platform with real-time dashboard and evidence packs.

EU AI Act Art. 6 risk classifier
Auto-generated DPIA + tech docs
Tamper-evident audit console
GitHub App for repo scanning

Comply Product

📚 50 Specs¶

Open specifications defining every protocol layer.

Core protocol, headers, envelope
Continuation, DPE, safety policy
Session tokens, dispatch, CKF
Conformance, security, zero-CKF
CDR/CDGR, CSO, STL, storage engine
Gateway, Comply, Scan products

Browse Specs

Who CRP is for¶

Developers¶

Ship governed AI in one line. Get context quality, safety scoring, and audit trails without building infrastructure.

Enterprises¶

EU AI Act, ISO 42001, GDPR compliance - auto-generated from real runtime evidence. SSO, data residency, private cloud.

CISOs & Compliance Officers¶

Real-time risk scoring, tamper-evident audit chains, and regulator-ready evidence packs. Not consultant PDFs.

AI Researchers¶

Compare strategies (CRP vs RAG vs Injection vs Hierarchical) with built-in benchmarking. Reproduce everything.

Startups¶

Free tier (100 calls/mo). Starter at $49/mo ($490/yr). Scale when you need SSO and data residency. No seat-based pricing.

Regulated Industries¶

Healthcare (HIPAA), finance (SOX), public sector. Air-gapped deployment, 7-year retention, signed DPA.

Read the full "Who is it for" breakdown

Before & After¶

❌ Without CRP¶

Task: "Write a 30-section K8s guide"
→ Output: 592 words, 8/30 sections
→ Truncated mid-sentence
→ No continuation
→ No quality score
→ Context lost between calls
→ No compliance evidence

✅ With CRP¶

Task: "Write a 30-section K8s guide"
→ Output: 6,993 words, 25/30 sections
→ Automatic multi-window continuation
→ Quality tier: A
→ Full provenance DAG
→ Knowledge persists in CKF
→ EU AI Act evidence generated

Same model. Same hardware. Same task. CRP produces 11.8x more content at identical throughput. The difference: CRP finishes the task - and proves it was safe.

Trusted by standards bodies¶

IETF submitted IANA submitted IEEE SA submitted ISO/IEC JTC 1/SC 42

Aligned with OWASP Top 10 (2025), ISO/IEC 42001:2023, EU AI Act (2024/1689), ISO/IEC 27001:2022, and NIST AI RMF.

Ready to govern your AI?

Start free. No credit card. 60-second setup.

Get Started Explore SDK

Get started¶

Choose your path. Every option includes the open SDK, HMAC audit chain, safety profiles, and control-evidence output for EU AI Act, AIUC-1, ISO 42001, and NIST AI RMF.

Quickstart Control-evidence mapping SDK Guide Self-hosting View Pricing

Position every agent on the right task,with the right context and tools. ¶

Controls are easy to claim. CRP proves they operate.¶

Outcomes & ROI¶

📝 Finish long tasks¶

⚖️ Cut compliance time¶

🛡️ Reduce hallucination risk¶

🚀 Deploy faster¶

Trusted by teams building with AI¶

Try it in 5 lines¶

What CRP solves¶

Context dies at 8 sections¶

No safety guardrails¶

Compliance is manual¶

Integration is expensive¶

Explore by topic¶

🛡️ AI Safety¶

⚖️ AI Governance¶

📋 AI Compliance¶

🧠 Context Management¶

🏅 Control Evidence¶

🚀 CRPv5 Roadmap¶

Capabilities¶

Unbounded context¶

6-stage extraction¶

13-stage DPE safety¶

Knowledge fabric¶

Two-sided provenance¶

Compliance evidence¶

One-line integration¶

Agent-safe¶

Positioned tool loop (v5)¶

The business case¶

Cost of unmanaged AI¶

Value of CRP governance¶

Three products. One agentic positioning engine.¶

🌐 CRP Gateway¶

📋 CRP Comply¶

🔍 CRP Scan¶

What's actually shipped¶

🐍 Python SDK¶

🖥️ CRP CLI¶

⚙️ GitHub Action¶

🌐 Gateway¶

📋 CRP Comply¶

📚 50 Specs¶

Who CRP is for¶

Developers¶

Enterprises¶

CISOs & Compliance Officers¶

AI Researchers¶

Startups¶

Regulated Industries¶

Before & After¶

❌ Without CRP¶

✅ With CRP¶

Trusted by standards bodies¶

Get started¶

Position every agent on the right task,
with the right context and tools. ¶