End AI Hallucinations.
Prove Every Citation Cryptographically.

Verifiable Documents adds a zero-knowledge proof layer to any document corpus.
Every AI answer comes with an independent, on-chain verifiable proof — without exposing a single page of your documents.

● 527 documents proven · ● Plonky2 ZK circuits · ● Live on Horizen EVM · ● zkVerify verification

No more fake citations. No more exposed IP. Just provable truth.

The Problem

AI systems hallucinate citations — confidently attributing claims to documents that don't exist. Traditional Retrieval-Augmented Generation (RAG) helps, but can't prove a result actually came from your corpus. Verifiable Documents fixes both.

🤖 AI Hallucinates Citations

ChatGPT confidently cites a Pentagon report that doesn't exist — sounding completely authoritative while being entirely fabricated. A search engine built on AI inherits this failure mode. Users have no way to tell fact from fiction.

🔍 Traditional RAG Has No Proof

Standard RAG retrieves relevant chunks from a corpus, but provides no cryptographic evidence that the chunks actually belong to the committed documents. It's a trust claim, not a proof — and it's useless if the corpus owner goes offline.

🌐 Exposing Documents Compromises IP

Sharing full documents to prove authenticity defeats the purpose for proprietary content. Publishers and researchers can't expose their corpus just to prove citations are real.

⛓️ Centralized Proofs Are Bought Off

A trusted third party that issues proof certificates is a single point of failure. If the authority is compromised or disappears, the proof is worthless.

Traditional RAG gives you answers. Verifiable Documents gives you provable truth.

🔐

Trust

Every citation is provably authentic. No faith required — only cryptography.

🛡️

Privacy

ZK proofs verify authenticity without revealing a single document page.

Speed

Proofs generate locally in milliseconds. On-chain verification in seconds.

How It Works

Verifiable Documents adds a cryptographic layer to document search. Every result carries a proof that can be independently verified — without revealing the underlying documents.

1

Commit

Upload your corpus once. The system builds a Poseidon Merkle tree over all chunks and anchors the root on Horizen EVM — permanently and publicly.

2

Search

Query the corpus semantically. Each result comes with a ZK proof generated locally in milliseconds — proving the passage belongs to a committed document.

3

Verify

Anyone verifies the proof on zkVerify — the result is provably authentic, without exposing your documents.

🔮
Zero-Knowledge Magic: The verifier learns a passage came from your corpus — but learns nothing about any other document in it. Privacy and proof, simultaneously.
Corpus Setup (one-time indexing)
📄
Your PDFs
Pipeline
(extract, chunk, embed)
🔓
Merkle Tree
+ ZK Proof
🖸
Horizen
Blockchain
Query (at search time)
🔍
Search
API
🔓
Generate
ZK Proof
🚈
Kurier
API
zkVerify
Verified
Result

Documents are anchored on-chain once. Every search triggers a fresh ZK proof verified via Kurier → zkVerify — provably, without revealing the documents.

See It Live: Military Manuals Demo

527 US Army field manuals and technical publications — fully searchable with cryptographic provenance. Every result includes a verifiable ZK proof that proves the answer came from the real document — no hallucinations, no trust required.

Try these powerful searches:
"rules of engagement urban operations"
✅ Verified Proof generated · 18s on zkVerify
FM 3-06 — Urban Operations
"...units must maintain situational awareness while distinguishing between combatants and civilians in close-quarters engagement..."
Provenance: FM3-06 · Page 4-12 · Chunk 7 · Merkle proof verified on Horizen
✅ 527 documents anchored ✅ On-chain Merkle root ✅ Plonky2 proofs ✅ Verified on Horizen + zkVerify

What You Can Build

Verifiable Documents applies anywhere documents, authenticity, and privacy need to coexist.

📚

Historical Archives

Publishers share citations from vintage manuals — peers verify authenticity without needing access to the original documents.

🔬

Academic Research

Researchers publish field notes and datasets with ZK proofs of origin. Peers verify provenance without accessing the full corpus.

⚖️

Legal & Compliance

Contracts and regulatory filings are timestamped on-chain. Auditors verify authenticity in seconds — without reading the sensitive terms.

📰

Investigative Journalism

Newsrooms publish document archives with cryptographic proof of origin — readers verify authenticity without the source having to release raw files.

🎨

Creative Works

Photographers, artists, and writers establish public timestamps of original works without publishing the work itself.

🏭

Supply Chain

Compliance certificates and authenticity reports prove a document is original and unmodified — without exposing proprietary process details.

🏛️

Public Records

FOIA releases and government documents carry verified timestamps. Citizens confirm authenticity without needing the full archive.

Set Up Your Own

Everything is open source. Clone the repo, configure your credentials, and deploy a ZK-Proven search system for your own document corpus.

Proof Layer

zkVerify

The proof verification layer on Horizen. Plonky2 proofs are submitted through Kurier's API and verified on-chain at zkVerify — low gas, high throughput. zkverify.io →

Blockchain

Horizen EVM

The MerkleRootRegistry contract anchors your corpus root permanently on Horizen — publicly verifiable, censorship-resistant. Requires a wallet with ZEN for gas. Foundry deployment scripts included.

API

Kurier

Submit ZK proofs to zkVerify through Kurier's API. Get an API key at kurier.xyz. Handles proof job lifecycle, retries, and status polling.

Data

Your Documents

Any PDF corpus you hold the rights to. The pipeline extracts text, runs OCR on scanned pages, chunks, embeds, and upserts to Qdrant — then builds the Merkle proof structure.

Full instructions, deployment scripts, and circuit binaries on GitHub.

Clone on GitHub →
CloneConfigureDeploy

Technology Stack

Built on battle-tested open-source infrastructure.

Horizen EVM

Decentralized EVM chain for on-chain Merkle root commitments. Public, censorship-resistant, and permanent.


horizen.io →

zkVerify

Zero-knowledge proof verification layer on Horizen. Off-chain computation, on-chain verification, low gas cost.


zkverify.io →

Kurier

API service for submitting and tracking ZK proof jobs to zkVerify. Handles the dev ops of proof submission so you don't have to.


kurier.xyz →

Plonky2

Fast recursive SNARKs for local proof generation. Pre-built circuit binaries included (depth 5–12). No ZK expertise required to run.


GitHub →

Ready to Make Your Documents Verifiable?

Turn any document corpus into a provably authentic, hallucination-resistant search engine.
Open source. Self-hosted. Running today on Horizen + zkVerify.

No trusted third parties. No exposed IP. Just provable truth.