Description

Signatrust

The trust, verification, and accountability layer for autonomous AI agents.


signatrust.net · Documentation · ADR Specification · Verify a receipt

—

What is Signatrust?

Signatrust generates and verifies cryptographically signed AI Decision Receipts. Each receipt is independently verifiable evidence of an autonomous AI action — without exposing prompts, outputs, or sensitive business data.

Every AI decision becomes a verifiable, non-repudiable fact.

—

Core capabilities

| Capability | Description |
|—|—|
| Ed25519-signed receipts | Every AI decision is sealed into a tamper-evident, hash-chained receipt |
| Decision Boundary Disclosure | Receipts declare which domains were evaluated and excluded — backed by versioned sector schemas |
| Independent verification | Any party can verify a receipt without accessing raw data or your systems |
| Chain of custody | Tool-call-level sealing with shared trace_id across agent steps |
| Privacy-preserving | Only SHA-256 fingerprints leave your infrastructure — never raw content |
| Sector schemas | 8 pre-built schemas (automotive, medical, credit, insurance, HR, legal, industrial IoT, generic) |
| AI Agent agnostic | Works with any model, framework, or orchestration platform |

—

Packages & integrations

| Directory | Description | Status |
|—|—|—|
| sdk-js/ | JavaScript/TypeScript SDK (signatrust on npm) | ✅ Live |
| n8n-node/ | Official n8n Community Node | ✅ Live |
| langchain-signatrust/ | LangChain integration | ✅ Live |
| crewai-signatrust/ | CrewAI integration | ✅ Live |
| autogen-signatrust/ | AutoGen integration | ✅ Live |
| openai-agents-signatrust/ | OpenAI Agents SDK integration | ✅ Live |

—

Quick start (JavaScript/TypeScript)

import { Signatrust } from 'signatrust';
const str = new Signatrust({ apiKey: process.env.SIGNATRUSTAPIKEY });
const { receipt } = await str.sign({
  model: { provider: 'openai', name: 'gpt-4o', version: '2026.4' },
  decision: {
    type: 'loan_rejection',
    input,            // hashed locally — raw data never leaves
    output,           // hashed locally
    risk_level: 'high',
    human_review: true,
    policies: ['eu-ai-act-high-risk'],
  },
});
const { valid } = await str.verify(receipt.id); // true

—

Decision Boundary Disclosure (DBD)

Not just what the AI decided — what it didn’t consider. DBD extends receipts with a signed scope declaration:

const { receipt } = await str.sign({
  decision: { type: 'collisionassessment', input, output, risklevel: 'high' },
  scope_declaration: {
    sectorschemaid: 'automotive_collision.v1',
    domains_evaluated: [
      { domainid: 'physicsimpact_force', status: 'computed' },
      { domainid: 'vehiclestructural_integrity', status: 'computed' },
    ],
    domains_excluded: [
      { domainid: 'occupantbiologicalimpact', reason: 'nodatasourceavailable', excludedby: 'datagap' },
      { domainid: 'pedestrianthirdpartyimpact', reason: 'notinscopeofthismodule', excludedby: 'design' },
      // ... all remaining domains must be declared
    ],
  },
});
console.log(receipt.scopedeclaration.coverageclassification); // 'partial'

Key rules:

• Every domain in the sector schema must be classified as evaluated or excluded — silent omission is rejected at signing time.
• Coverage classification (full/partial/minimal), disclosure statement, and schema hash are computed server-side.
• The scope declaration is inside the signed body — non-repudiable.

Browse schemas: GET https://signatrust.net/api/v1/schemas

—

Chain of custody — tool-call-level sealing

const trace = str.trace({ mode: 'everytoolcall' });
await trace.step({ steptype: 'toolcall', toolname: 'search', decision: { type: 'searchresult', output: results } });
await trace.step({ steptype: 'toolcall', toolname: 'dblookup', decision: { type: 'lookup', output: rows } });
await trace.step({ steptype: 'finalresponse', decision: { type: 'loandecision', input, output, risklevel: 'high' } });
console.log(trace.traceid, trace.receipts.length); // trace..., 3

—

How it works

AI Agent makes a decision
        ↓
POST /api/v1/receipts (with optional scope_declaration)
        ↓
Returns: receipt + signature + hash + verify_url
        ↓
Anyone verifies at: https://signatrust.net/verify

No raw prompts, no model outputs, no sensitive data leaves your system.

—

Links

• Platform: signatrust.net
• Documentation: signatrust.net/docs
• ADR Specification: signatrust.net/adr
• Sector schemas: signatrust.net/api/v1/schemas
• Verify: signatrust.net/verify
• Enterprise: partners@signatrust.net

—

© 2026 Signatrust — Apache-2.0