ProofLink SDK

Every other AI-accountability standard is a PDF. ProofLink is a running ledger of 79,000+ cryptographically-sealed AI actions you can verify right now — not a spec, a live chain. → verify.itechsmart.dev

Seal + verify ProofLink Receipt Standard v3.0 receipts from Python or TypeScript. Every autonomous action gets a receipt: SHA-256 hash-chained, Ed25519-signed, Bitcoin-anchored, publicly verifiable. Don't trust the AI — trust the math.

Framework examples

Drop-in accountability for your stack — each uses a published SDK and the live public ledger. See examples/:

Example	Stack
FastAPI	verify endpoint + @accountable seal decorator
Next.js	/api/verify/[id] route + live-integrity server component
LangGraph	seal a receipt for every agent node, then verify the trail
CrewAI	task_callback seals a receipt per completed task
Claude Desktop	MCP config — 65 tools, verify receipts from Claude

Not a spec — a running chain

Live snapshot (2026-07-02, from /api/chain + /api/stats): 79,000+ receipts, chain intact (chain_intact: true, 0 breaks), 2,100+ strict cryptographically-verifiable v3 receipts (and every new action is sealed as v3), 13,700+ Bitcoin-anchored (~17%, growing daily). Verify any of it yourself with the 5-line snippet below — no iTechSmart account, no trust required.

Honest two-era note. v3 receipts (schema_version "3.0") are strict and fully cryptographically verifiable (hash + canonical re-derivation + Ed25519 + chain link). Legacy v1/v2 receipts are pointer-linked and preserved unmodified — disclosed openly at /api/stats. We do not claim all 79k are strict-verifiable; 2,100+ v3 are, and the count grows with every action.

Built for the regulations

Regulation / framework	ProofLink field / mechanism that satisfies it
EU AI Act (Reg. 2024/1689) Article 12 — automatic tamper-evident logging for high-risk AI	Append-only hash chain; every seal() records timestamp, actor, action, subject, outcome, details
NIST AI RMF 1.0 — MEASURE 2.7 / MANAGE 4.1 — monitoring evaluated & documented	security / platform_fix / platform_health_check receipts, signed & immutable; actor distinguishes system vs. agent vs. operator
CMMC L2 — AU.L2-3.3.1 / AU.L2-3.3.8 — retain & protect audit logs	SHA-256 chain + Ed25519 make any edit/deletion/reorder detectable; Bitcoin anchoring adds external existence proof
SOC 2 — CC7.2 / CC7.3 / CC8.1 — anomaly monitoring & change management	signal_classified / security receipts; config_change records {before_hash, after_hash, diff_summary}
ISO/IEC 42001:2023 — Clause 9.1 — retain documented monitoring evidence	The receipt ledger is the retained cryptographic evidence; compliance_tags seal the control claim inside the signature

Connect anything — every call seals a receipt

• MCP server — verify/search receipts from any MCP client (Claude, Cursor, Copilot, LangGraph, CrewAI): prooflink_verify_receipt, prooflink_search_receipts, prooflink_verify_chain.
• FastAPI / REST — verify.itechsmart.dev exposes /api/export, /api/verify/<id>, /api/chain, /api/stats, /api/anchors, /api/how-to-verify.
• This SDK — Python (pip install prooflink # NOT YET PUBLISHED to PyPI — build from this repo (see below)) and TypeScript (npm install @itechsmart/prooflink) wrap seal + verify with idiomatic bindings.

Related: the ProofLink Receipt Standard (spec + conformance suite) and the zero-dependency prooflink-verifier. ProofLink aligns conceptually with the IETF Internet-Draft draft-sharif-agent-audit-trail-00 (same problem, shared hash-chain core) while differing deliberately on canonicalization (json.dumps, not RFC 8785 JCS) and signature (Ed25519, not ECDSA P-256) — see §11 of the standard.

What it does

ProofLink turns "the agent did X" into a receipt you can prove later. Each receipt is:

• Hash-chained — every receipt links to the previous one's SHA-256
• Bitcoin-anchored — submitted to 4 OpenTimestamps calendars, settled in the next Bitcoin block
• Publicly verifiable — anyone can replay the chain at verify.itechsmart.dev
• Tamper-evident — modifying a past receipt breaks every receipt that follows

The SDK is a thin wrapper. The cryptography lives in append.py (the canonical seal logic) and the verify API. The SDK adds idiomatic language bindings — it does not reimplement.

Two modes

Mode	Available methods	Where it runs
Server (append.py present)	seal, verify, chain_status, submit_to_ledger	iTechSmart UAIO host or any host with /opt/itechsmart/audit_ledger/append.py
Client (verify API only)	verify, chain_status	Anywhere with HTTPS access to verify.itechsmart.dev

seal() raises a clear error in client mode — read-only methods still work.

Install

Python

pip install prooflink

Node

npm install @itechsmart/prooflink-verifier   # published verifier (v2.x). The full @itechsmart/prooflink SDK is not yet on npm — build from this repo.

Quick start

Conforms to the ProofLink Receipt Standard v3.0.

Python — verify a receipt (5 lines)

verify() reproduces the live verification exactly: SHA-256 hash recompute, canonical re-derivation, Ed25519 signature, and (optionally) the prev_hash chain link. It returns True only if all checks pass.

import prooflink
receipt = prooflink.fetch("c58347c60394a21f")   # pull any receipt from the public API
assert prooflink.verify(receipt)                # cryptographically verify it — True

Want the per-check breakdown? Use verify_receipt():

result = prooflink.verify_receipt(receipt, prev_hash="<previous entry hash>")
# {'valid': True, 'id': 'c58347c60394a21f', 'schema_version': '3.0',
#  'checks': [{'name': 'hash_integrity', 'passed': True, ...}, ...], 'errors': []}

Python — seal a receipt (server mode)

seal() shells out to the canonical append.py (only available on a host that has /opt/itechsmart/audit_ledger/append.py + SEAL_TOKEN):

import prooflink
r = prooflink.seal("restarted suite-nginx after OOM kill",
                   category="container_restart", actor="system:supervisor",
                   subject="suite-nginx", outcome="healthy 12s after restart")
print(r["id"], r["hash"])   # short form; verify with prooflink.fetch(r["id"])

Node / TypeScript — verify a receipt (5 lines)

See typescript/ for the TS verifier. It reproduces the same four checks using Node's built-in crypto (no runtime dependencies):

import { verify } from "./src/index";
const res = await fetch("https://verify.itechsmart.dev/api/verify/c58347c60394a21f");
const { receipt } = await res.json();
console.assert(verify(receipt));   // true

verify()/verify_receipt() require the cryptography package (the same library as the published reference verifier). seal(), verify_chain(), recent(), and stats() are stdlib-only.

Receipt schema

Every receipt has the same shape, regardless of language binding:

{
  "id": "437f2bbd7fb221ac",
  "timestamp": "2026-06-02T22:14:08.231054+00:00",
  "category": "container_restart",
  "actor": "system:supervisor",
  "subject": "suite-nginx",
  "action": "restarted after OOM kill",
  "outcome": "service healthy 12s after restart",
  "details": { "pid": 12345, "oom_score": 800 },
  "hash_sha256": "437f2bbd7fb221ac7ce8ff917f84135e7607c5f5b1282354f4c4ac6e0ef8560b",
  "prev_hash": "<previous receipt's hash_sha256>",
  "tamper_detected": false,
  "human_input": false,
  "auto_resolved": true
}

seal() returns the short form {ok, id, hash}. Full receipts are returned by verify().

Configuration

Param	Default	Purpose
append_py	/opt/itechsmart/audit_ledger/append.py	Path to the canonical seal CLI
verify_url	https://verify.itechsmart.dev	Base URL of the verify API
python_bin	python3	Python interpreter for the seal subprocess
timeout	30s	Per-seal subprocess timeout
no_ots	False (per-call)	Skip Bitcoin anchoring (~70x faster seal; receipt still hashed and chained)

Status of the verify API

Endpoint	Status
GET /api/chain	✅ Live — returns {chain_intact, total, breaks}
GET /api/receipts?hash=...	⚠ Currently returns HTML — JSON response in progress (sprint item H3)
GET /api/stats	⚠ Same as above

Until H3 lands, verify(hash) may return parsed HTML scaffolding rather than the receipt JSON. chain_status() is the production-stable read path.

License

MIT — see LICENSE.

Author

iTechSmart Inc. — djuane@itechsmart.dev