@dgp-standard/dgp-js

Reference implementation of the DeAlgo Governance Protocol (DGP) v1.0

✅ DGP v1.0 Conformant — Passes all 8 canonical protocol vectors with bit-for-bit accuracy.

Protocol Spec: DeAlgo Governance Protocol (DGP) v1.0

Overview

The Governance SDK enables inference-time governance across multiple LLM vendors (OpenAI, Google, Anthropic) without fine-tuning. Control AI behavior through portable JSON constraint capsules that enforce:

• RFE (Reflection-First Execution): Plan before acting
• SPS (Stability-Preserving Steps): Prefer safe operations
• SEG (Scope-Enforcing Gates): Respect task boundaries
• FOP (Founder-Oversight Protocol): Escalate high-risk actions

Quick Start

Reference Implementation (evaluates LLM output against governance policy):

npm install @dgp-standard/dgp-js

import { GovernanceEngine } from './src/core/GovernanceEngine.js';

// Create engine with deterministic time injection
const engine = new GovernanceEngine({ 
  now: () => "2026-02-11T00:00:00.000Z", 
  engineVersion: "1.0.0" 
});

// Evaluate LLM output against capsule policy
const report = engine.evaluate({ 
  capsule,   // Governance policy (JSON)
  task,      // Task context (scope, risk level)
  output,    // LLM-generated text to evaluate
  baseline   // Optional: previous output for drift comparison
});

// Use verdict to gate/retry/escalate/block
console.log(report.verdict);        // "ALLOW" | "RETRY" | "ESCALATE" | "BLOCK"
console.log(report.analysis.score); // 0-100 compliance score
console.log(report.violations);     // Detected violations (if any)
console.log(report.actions);        // Recommended actions

See DGP.md for protocol overview and integration patterns.

Protocol Conformance

This SDK is the reference implementation of the DeAlgo Governance Protocol (DGP) v1.0.

Validate conformance:

node --test test/contract-compliance.test.js

Expected: 8/8 canonical vectors passing (bit-for-bit match).

Protocol artifacts:

• Spec (RFC): ENGINE_CONTRACT.md
• Conformance Suite: canonical-v1.json (8 normative test vectors)
• Conformance Guide: CONFORMANCE.md

DGP positioning: OAuth + OpenAPI + PCI for AI governance

Features

• ✅ Multi-vendor support: OpenAI GPT, Google Gemini, Anthropic Claude
• ✅ Zero fine-tuning: Governance via prompt injection, not training
• ✅ Portable constraints: Single JSON spec works across all models
• ✅ Production-ready: Retry logic, rate limiting, error handling
• ✅ Extensible: Custom validators, capsules, and providers

Documentation

• Quick Start Guide
• API Reference
• Custom Capsules
• Framework Integration

Research

Based on peer-reviewed research validating portability across GPT-4o-mini, Claude Sonnet 4.5, and Gemini 2.0 Flash:

Paper: Inference-Time Governance via Portable Constraint Capsules
Validation: 2/3 models achieved ≥30% drift reduction, 3/3 achieved perfect scope compliance

Examples

See examples/ for:

• Basic usage with single provider
• Custom capsule creation
• Multi-provider comparison
• Framework integrations (Express, Next.js, LangChain)

License

MIT © DeAlgo Labs

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Questions? Open an issue or join our Discord