Ask most AI agent builders what their audit trail looks like, and you’ll get an uncomfortable pause. The honest answer is usually: a chat log. The agent read a contract, extracted some clauses, flagged a discrepancy, and moved on. But can you prove which version of the document it actually saw? Can you prove nobody touched the extracted text afterward? With most setups, no.
DocImprint is built specifically to close that gap. It gives AI agents cryptographically grounded document memory: every stored extract becomes a tamper-evident evidence bundle with a SHA-256 manifest, Merkle citation proofs, and an EIP-191 signature (a standard wallet signature format) you can verify entirely offline. DocImprint attests what was captured at extract time; you — or an auditor — can prove that attestation has not been altered since, without calling their API.
The product just launched on Product Hunt and it’s already attracting attention from agent builders working in legal, finance, and compliance. The core pitch is simple: if your agent acts on a document, you should be able to prove exactly what it read and when.
Key Features of DocImprint
Tamper-Evident Evidence Bundles
Every stored extraction produces a ZIP bundle. Typical contents: manifest.json (per-artifact SHA-256 hashes), signature.json (EIP-191 / secp256k1 signing metadata), chunks.merkle.json, page.md, screenshot.png, and ocr.txt — plus page.pdf when the original PDF is retained. Artifacts depend on source and options; not every file appears in every bundle. The manifest lists per-artifact hashes. Anyone can download the ZIP and recompute those hashes locally to confirm nothing changed since capture. That’s the whole point: the proof travels with the document, not with DocImprint’s database.
Merkle Citation Proofs
This is where it gets genuinely interesting for legal and compliance workflows. DocImprint chunks documents into paragraphs and builds a Merkle tree, with the root signed into the manifest. When an agent cites a specific clause, you can call verify-citation and get a Merkle proof back for that exact chunk. You can prove a specific paragraph came from a specific PDF without needing to expose the whole document.
One-Call Extraction via POST /v1/extract
Without DocImprint, an agent typically chains four separate tool calls: screenshot, OCR, parse, then hash manually. DocImprint collapses all of that into a single POST. The response comes back with a bundle_id, a manifest_sha256, and structured artifacts. One call, one payment, full provenance. Modes include summarize, qa, claim-check, extract-structured, and compare (plus lean extract / invoice / translate / describe).
MCP-Native and x402 Pay-Per-Call
DocImprint is built to fit inside agent workflows without friction. It ships an MCP server with namespaced tools — document.extract_text, url.extract, document.check_claims, bundle.verify, and receipt.verify — available in Claude Desktop, Cursor, and any MCP-compatible host. For autonomous agents that should not need a human-provisioned API key, DocImprint supports x402 (HTTP-native USDC micropayments) on Base. The agent signs a per-request payment; the paying wallet becomes the bundle owner. EIP-191 signing matches that wallet-native payment path: one crypto shape for pay and prove.
Collections for Cross-Document Q&A
Single bundles handle one document at a time. Collections let you group multiple bundles into a searchable corpus. An agent can then run semantic search or ask cross-document questions. This is particularly useful for legal matters or fiscal year audit reviews where you need answers spanning dozens of related filings.
How a Compliance Agent Actually Used This
Marcus, a backend developer at a mid-size fintech, was building a reconciliation agent to process supplier invoices. It was 2:30 in the afternoon on a Tuesday and he was trying to wire up an audit-ready pipeline before a Friday deadline. The problem: his LangGraph agent could extract invoice totals fine, but there was no way to prove to the compliance team which version of a PDF it had actually read, or that the extracted line items hadn’t been modified downstream.
He connected DocImprint via the MCP config in about five minutes, pasting the server block into his Cursor settings. Then he pointed his agent at a supplier invoice PDF using the extract_text tool. The response came back with a bundle_id and a manifest_sha256. He ran the claim-check mode next, validating the invoice totals against ERP records. Two line items flagged as contradicted, with quoted passages and Merkle proofs pointing to the exact paragraphs in the original PDF.
He downloaded the evidence bundle ZIP, unzipped it locally, and recomputed the SHA-256 hash on manifest.json using a single terminal command. It matched. No API call required at that point. He handed the bundle_id citations to the compliance team and they could independently verify the extraction had not been altered. The Friday deadline was met with actual proof, not just a log entry.
How DocImprint Processes a Document
Step 1: Send a single POST request
You send a URL or upload a PDF to POST /v1/extract with your chosen mode. For stored bundles, the API returns a 202 and a job_id. For smaller documents, you can add ?sync=true to get results inline.
Step 2: Receive the evidence bundle
Once processing completes, the bundle contains a manifest.json listing every artifact’s SHA-256 hash, a signature.json with the EIP-191 signing metadata, a Merkle chunks file, plus the actual PDF, screenshot, Markdown, and OCR text. All artifacts are hashed and signed together.
Step 3: Verify offline, free, any time
Call GET /v1/extract/{id}/verify to trigger a deep server-side recompute, or download the ZIP and recompute hashes locally. Verification and download are always free. The signing key is publicly available at GET /v1/keys, so the proof never depends on DocImprint being reachable or trustworthy at review time.
Pricing
DocImprint offers three access paths:
- Free tier: 100 credits per month, no credit card needed. Enough for 5 full bundles or 33 lean extract calls.
- x402 pay-per-call: USDC on Base, no account required. Prices range from $0.010 for lean text extracts up to $0.075 for a standard full bundle (up to 25 pages). Verify and download are always free.
- Pro API key: $29 per month via Stripe. Includes 555 full bundles or 3,333 lean calls per month, making the effective bundle cost $0.0523 versus $0.075 on x402.
Who Should Try DocImprint
If you are building agents that touch contracts, invoices, regulatory filings, or any document where someone might later ask “what exactly did the agent read?”, DocImprint fills a gap that logging tools simply cannot. It is especially well suited for legal hold workflows, SOX audit pipelines, and compliance agents where provenance is not optional. Start with the free tier at docimprint.com and run the quickstart in under five minutes.
