Regain Hermes™

The industry-standard protocol for safe, auditable, and traceable clinical AI supervision.

Overview

Regain Hermes™ is a protocol for clinical AI supervision. It defines the Epistemological Contract required for safe communication between reasoning agents (the "Brain") and independent supervision systems (the "Shield").

By standardizing how AI systems propose interventions, provide evidence, and measure uncertainty, Hermes enables a modular ecosystem of trusted clinical AI that is regulatory-ready from day one.

Read our Project Vision to learn how we are building the "Grammar of Safety."

Package Features

This package provides the reference implementation of the Hermes protocol:

• TypeScript Types - Complete type definitions for all Hermes messages.
• JSON Schema Validation - Runtime validation for inter-agent communication.
• Epistemology Utilities - Hard2Vary™ (HTV) scoring, uncertainty calibration, and evidence grading.
• Audit Tooling - Standardized formats for regulatory-grade, de-identified audit exports.
• Test Fixtures - A comprehensive library of clinical scenarios for unit testing.

Project Status: Public Alpha

Regain Hermes™ is currently in Public Alpha. The core protocol schemas and validation logic are stable for early adoption, but the broader commercial ecosystem (Certification and Managed Services) is currently under development.

Installation

# npm
npm install @regain/hermes

Certification (Roadmap 2026)

To ensure future ecosystem trust, we are developing a Certification Program with three proposed tiers:

• Hermes Compatible (Free) - Automated conformance.
• Hermes Certified (Paid) - Technical review & support.
• Hermes Certified Clinical (Paid) - Full clinical audit & regulatory readiness.

If you are interested in becoming a launch partner for the certification program, please contact [email protected].

Enterprise & Managed Services (Coming Soon)

For organizations that prefer a fully managed solution, Regain, Inc. is developing hosted supervision infrastructure:

• Regain Popper™ Cloud - Managed policy engine with deterministic safety gates and audit logging.
• Enterprise Support - SLA-backed support for mission-critical clinical deployments.
• Regulatory Packages - Pre-configured compliance documentation for FDA, HIPAA, and IMDRF.

These services are currently in development. Contact [email protected] to join the early access waitlist.

Documentation

For Developers

• Developer Guide - Step-by-step integration guide with workflows
• API Reference - Complete function and type documentation
• Epistemology Guide - HTV scoring and uncertainty deep dive
• Testing Guide - How to test your integration
• Protocol Specification - Full schema spec

Project Governance

• Project Vision - Our strategy for safe clinical AI.
• Certification Program - How to certify your implementation.
• Governance Model - How the standard is maintained.
• Contributing - Technical guidelines and our CLA.
• Trademark Policy - Usage guidelines for our marks.

Quick Start

Validate Messages

import { validateHermesMessage, parseHermesMessage } from '@regain/hermes';

// Option 1: Check validity without throwing
const result = validateHermesMessage(payload);
if (!result.valid) {
  console.error('Validation errors:', result.errors);
}

// Option 2: Parse with type safety (throws on failure)
const request = parseHermesMessage(payload);
console.log(`Trace: ${request.trace.trace_id}`);

Compute Epistemic Scores

import { computeHTVScore, getHTVQualityLevel } from '@regain/hermes';

const score = computeHTVScore({
  interdependence: 0.9,
  specificity: 0.85,
  parsimony: 0.8,
  falsifiability: 0.9,
});

console.log(score.composite);  // 0.8625
console.log(getHTVQualityLevel(score.composite));  // 'good'

Type Guard Pattern

import { isValidHermesMessage, type SupervisionRequest } from '@regain/hermes';

if (isValidHermesMessage(payload)) {
  // TypeScript knows payload is HermesMessage
  if (payload.message_type === 'supervision_request') {
    handleRequest(payload as SupervisionRequest);
  }
}

API Summary

Validation

| Function | Description | |----------|-------------| | validateHermesMessage(msg) | Validate against JSON Schema, returns { valid, errors } | | parseHermesMessage(msg) | Validate and parse, throws HermesValidationError on failure | | isValidHermesMessage(msg) | Type guard for Hermes messages |

Epistemology Utilities

| Function | Description | |----------|-------------| | computeHTVScore(dims, weights?) | Calculate Hard2Vary™ composite score (0.0-1.0) | | meetsHTVThreshold(score, threshold?) | Check if score meets minimum threshold | | getHTVQualityLevel(score) | Get quality level: 'excellent', 'good', 'moderate', 'poor', 'refuted' | | computeUncertainty(inputs) | Calculate calibrated uncertainty with drivers | | isUncertaintyAcceptable(unc, maxLevel?) | Check if uncertainty is acceptable | | compareEvidenceGrades(a, b) | Compare evidence strength (negative if a stronger) |

Builders

| Function | Description | |----------|-------------| | htvScore() | Fluent builder for HTV scores | | createFalsificationCriteria(...) | Create falsification criteria for claims | | createUniformHTVScore(value) | Create HTV with all dimensions equal |

See API Reference for complete documentation including constants, types, and examples.

Contributing

Contributions are welcome! Please read our contributing guidelines and governance model before submitting pull requests.

Trademarks

"Regain Hermes™", "Regain Popper™", "Popper™", "Regain Deutsch™", and "Hard2Vary™" are trademarks of Regain, Inc. See our Trademark Policy for usage guidelines.

Links

• GitHub Repository
• npm Package
• Issue Tracker
• Code of Conduct

Regain Hermes™, Regain Deutsch™, Regain Popper™, Popper™, and Hard2Vary™ are trademarks of Regain, Inc. All rights reserved.