@fuzentry/sdk-ts

TypeScript SDK for the Fuzentry AI governance platform.

Fuzentry enforces compliance policy before your AI calls execute — not after. Every execution produces an immutable, cryptographically signed audit record.

Access: This package is private. Install requires an npm token issued with your Fuzentry subscription.

Installation

npm install @fuzentry/sdk-ts

Set your credentials:

export FUZENTRY_API_KEY=your_api_key
export FUZENTRY_BASE_URL=https://api.tailoringtechforward.net

Quick Start

import { FuzentryClient, GovernanceError } from '@fuzentry/sdk-ts';

const fuzentry = new FuzentryClient({
  apiKey: process.env.FUZENTRY_API_KEY,
});

try {
  const result = await fuzentry.execute({
    surface: 'MODEL',
    context: {
      tenantId: req.tenant.id,        // from your JWT middleware
      folderId: req.body.accountId,   // from your request body
      actorId:  req.user.sub,         // from your JWT sub claim
      intent:   'loan-decision',      // declared once at the call site
    },
    input: applicantData,
    handler: (input) => callOpenAI(input),
  });

  return res.json(result.output);

} catch (err) {
  if (err instanceof GovernanceError) {
    // Policy denied — audit record already written
    return res.status(403).json({ denied: true, reason: err.reason });
  }
  throw err;
}

Core Concepts

ExecutionContext — 4 Required Fields

| Field | Source | Changes per request? | |---|---|---| | tenantId | JWT claim | Yes | | folderId | Request body | Yes | | actorId | JWT sub | Yes | | intent | Hardcoded at call site | No |

intent is declared once when you write the wrapper function. It is the governance label for what that code does — never computed from request data.

Surfaces

Every AI operation is governed at one of seven surfaces: AUTH → PROMPTS → MODEL → MEMORY → TOOL → PLUGIN → EXPORT

Enforcement Modes

| Mode | Behavior | |---|---| | OBSERVE | Policy evaluates, violations logged, execution proceeds | | SHADOW | Policy evaluates, hypothetical denial proof generated, execution proceeds | | ENFORCE | Policy evaluates, violations block execution, audit record written |

API Reference

fuzentry.execute(params)

Execute a governed AI operation. Handler only invoked on ALLOW. Audit record written regardless.

fuzentry.queryAuditRecords(params)

Query governance decisions by tenant, folder, intent, or surface.

fuzentry.getDenialProof(auditRecordId)

Retrieve a cryptographic denial proof for compliance evidence.

fuzentry.ping()

Verify API key and tenant connectivity. Call this during integration setup.

Support

For design partner integration support: [email protected]