Live · hivemorph.onrender.com/v1/verify Patent Pending

XCALIBUR VERIFY

Zero-knowledge proof that every piece of context injected into your AI prompt is a legitimate member of your corpus. Verifiable by you. Verifiable by your regulator. Trust the math, not us.

ML-DSA-65 · FIPS 204 Merkle corpus commitment Tenant-isolated Live demo below Patent Pending
The question: "Did that context actually come from my corpus?"  The answer: a proof you can verify without trusting anyone.
Live pipeline

VERIFY rides on top of CCACW

The fourth leg of XCALIBUR. Sits after AMPLIFY, before the response leaves the rail. Every injected corpus entry gets a Merkle inclusion proof. The whole certificate is signed under a post-quantum key.

CALLER
COMPILE
COMPRESS
LLM
AMPLIFY
VERIFY
CALLER + cert
Corpus Merkle tree · proof path highlighted
ROOT · SIGNED H(A·B) H(C·D) ENTRY A ENTRY B ENTRY C ENTRY D --- proof path · Entry A injected ---
Verify certificate · live values populate on run
version"xcalibur-verify/v1"
commitment_rootclick Run Verification →
commitment_ver
proof_mode"standard"
leaf_hash
siblings[]
original_hash
injected_hash
fidelity_score
signature
valid
Calls POST /v1/verify/demo/run on hivemorph.onrender.com · real Merkle, real signature
What VERIFY proves

Three claims, one certificate

01 · Corpus membership

The injected context is yours

The context injected into your prompt is a verified member of your committed corpus. Not fabricated. Not substituted. Provably yours.

02 · Zero contamination

No tenant can spoof your proof

The cryptographic boundary between tenants is enforced at proof generation. No other tenant's corpus can produce a valid proof against your commitment root.

03 · Content intact

Nothing slipped in or out

Your original prompt content is hashed and bound into the certificate. The SHA3-256 commitment proves nothing was inserted, replaced, or modified before the LLM.

Why nobody else has this

Logging is not a proof

Azure OpenAI · AWS Bedrock · Vertex

Attestation is a logging system. You trust their audit trail. You cannot independently verify corpus membership. Your regulator audits them — you don't verify the math.

No proof · trust required

LlamaIndex · LangChain · AutoGen

RAG frameworks inject context with no attestation layer. What went into the prompt is whatever the framework retrieved. No commitment, no proof, no receipt.

No attestation

EAS · Sismo · Galxe Passport

On-chain attestation of identity claims. Not inference context. Not corpus membership. Not AI prompt integrity. Different problem, different layer.

Wrong layer

XCALIBUR VERIFY

Merkle-committed corpus. ML-DSA-65 signed root. Per-call inclusion proof. Caller-verifiable at public endpoint. No trust required. Post-quantum hardened. Patent Pending.

Only this exists
Public API · live now

Five endpoints, one base

All paths under https://hivemorph.onrender.com. The proof verification endpoint is unauthenticated — auditors hit it directly.

GET/v1/verify/statusService health, signature algorithm, demo corpus size
POST/v1/verify/commitment/buildBuild a signed Merkle commitment for a tenant's current corpus
GET/v1/verify/commitment/{tenant_id}/currentPublic lookup of a tenant's current signed commitment
POST/v1/verify/proof/generateInclusion proof for an entry — standard or zk mode
POST/v1/verify/proof/verifyVerify proof + commitment without trusting the server
POST/v1/verify/disputeCaller dispute with 24h resolution against retained corpus snapshot
POST/v1/verify/demo/runEnd-to-end pipeline run — what the button on this page calls
LIVE · hivemorph.onrender.com/v1/verify XCALIBUR umbrella → AmpliHive widget → Pricing → Patent Pending