What is the Context Recall Protocol?

CRP is a cryptographic middleware layer that sits between an AI agent and its storage backend. It verifies that every piece of context the agent retrieves is bit-for-bit identical to what was stored, records each retrieval as a tamper-evident ledger event, and can prove that all context used in any given decision is known and unmodified.

What happens when tampered context is detected?

The content is quarantined before reaching the agent's reasoning pipeline. A dual-hash tamper evidence record is anchored to the ledger — preserving both the original hash and the modified-content hash as proof. The agent receives an error code. The content does not proceed.

What is Decision Completeness?

For any specific decision an agent makes, CRP can produce a cryptographic proof that it has captured every context retrieval during that decision process — nothing was hidden, bypassed, or injected. This is the proof regulators, insurers, and auditors need when they ask 'what did the agent know?'

Does the context content go on the ledger?

No. The Hedera record contains only the recall ID, enrollment ID, hash match flag, agent ID, and timestamp — not the content itself. Any third party can verify that a recall occurred and whether the content was authentic, without seeing what was recalled.

CRP — Context Recall Protocol | Cryptographic proof of what context an AI agent had at the moment of any decision | Context is hash-sealed at decision time — cannot be reconstructed or altered after the fact | What gets recorded: Full context window hash, model identifier, token count, timestamp, session chain position, operator policy version in effect | Auditors can independently verify the agent's information environment without querying the originating system | Closes the "what did it know?" gap in enterprise AI compliance | Core metric = Decision Context Completeness | Patent filed 2026-04-18 | ← Full Platform

Layer 02 · Context Recall · Patent Pending

What did the agent know
when it decided?

Know what your agent saw before it acted, and prove it to anyone who asks — auditor, regulator, or customer. We call this the Context Recall Protocol (CRP): it verifies every piece of context your agent retrieves, catches tampering the moment it happens, and anchors proof on Hedera. Independently auditable. No cooperation required.

Tamper Detection at Recall Hedera-Anchored Proof Decision Completeness Storage-Agnostic

Talk to us See How It Works

How It Works

Enroll. Verify. Prove.

Three stages. Every context retrieval is verified at the moment that matters — recall time, not just storage time.

01 — Enroll

Hash it before it's stored

When context is written, CRP computes a cryptographic hash and anchors it to Hedera before the write completes. The reference exists on the ledger before the data can ever be touched.

02 — Verify

Check it at every recall

When the agent retrieves context, CRP recomputes the hash from what's actually returned and compares it to the anchored reference. Match — delivered. Mismatch — quarantined, evidence record created, agent notified. Never silently passed through.

03 — Prove

Complete decision audit trail

For any agent decision, CRP produces a cryptographic proof that it captured every context retrieval during that process — nothing hidden, nothing bypassed, nothing injected. Any auditor can verify it independently.

What You Get

What auditors and regulators ask for.

🔒

Context Injection Defense

Modified context is detected and quarantined before reaching the agent's reasoning pipeline. Both the original and modified hashes are preserved as ledger evidence.

📋

Per-Recall Attestation

A receipt for every context pull. Each retrieval generates a signed, ledger-anchored record. Any third party can verify: did this agent recall this context at this time, and did the hash match?

✅

Decision Completeness

Prove that all context used in any specific decision is accounted for. No gaps, no bypasses. The full reasoning environment, cryptographically preserved.

🗄️

Works On Your Stack

CRP operates above any storage backend — vector databases, SQL, key-value stores, file systems. No changes to your infrastructure required.

⚡

Real-Time Enforcement

Verification happens at retrieval time, not in a batch audit. Tampered content never reaches the agent. There's no window where bad data could influence a decision.

🔗

No Platform Dependency

All proofs are anchored to Hedera. Auditors verify against the public ledger — not against your platform's logs. No cooperation required from the operator being audited.

Where It Fits

Layer 02 of the accountability stack.

Every layer builds on the one before it. CRP sits above Flash Tag identity and beneath the enforcement and scoring layers.

Layer 01

Flash Tag

Who is the agent and who deployed them.

Layer 02

Context Recall

What the agent knew when it decided.

Layer 03

Governance

What the agent was allowed to do.

Layer 04

Trust Scoring

How trustworthy is its behavior over time.

Layer 05

Commerce

How transactions are committed and settled.

Layer 06

Lifecycle Registry

The complete record from birth to closure.

Questions

Common questions.

Today, when an AI agent makes a decision, there's no way to prove what context it used. Anyone can claim different context was retrieved, or that context was modified between storage and recall. CRP closes that gap — every recall is verified at retrieval time, signed by the agent, and permanently recorded on Hedera.

The content is quarantined before reaching the agent. A dual-hash tamper evidence record is anchored to Hedera — preserving both the original hash and the modified hash as proof that tampering occurred, without disclosing the content. The agent receives an error. The tampered data goes nowhere.

No. The Hedera record contains only the recall ID, hash match result, agent ID, and timestamp. Not the content. Any auditor can verify a recall occurred and whether it was authentic — without seeing what was recalled.

CRP is patent pending. Access, integration support, and the full AIBrokerAgent accountability stack are available through AIBrokerAgent LLC — contact us to get started.

What did the agent knowwhen it decided?