Governance infrastructure for autonomous AI agents.
Molt is a monorepo containing three packages that form a complete governance stack for AI agent systems: identity verification, policy enforcement, and inter-agent communication.
| Package | Description | Version | Tests |
|---|---|---|---|
@molt/captcha |
Reverse CAPTCHA — SMHL challenge engine for AI verification | 1.0.0 | 51 |
@molt/permit |
Cedar-based policy engine with SQLite audit and JIT tokens | 0.1.0 | 71 |
@molt/mesh |
Agent interoperability bus with federation and exchange | 0.1.0 | 467 |
@molt/eval |
Agent evaluation engine — metrics, regression detection, release gating | 0.1.0 | 108 |
molt |
Meta-package that re-exports all four | 0.1.0 | — |
Total: 697 tests
# Install all dependencies
npm install
# Build all packages
npm run build
# Run all tests (697 tests across 4 packages)
npm testnpm install moltimport { MoltCaptcha, MoltPermit, MoltMesh, MoltEval } from 'molt';
// Or use namespaced imports
import { captcha, permit, mesh, eval } from 'molt';npm install @molt/captcha
npm install @molt/permit
npm install @molt/mesh
npm install @molt/eval┌──────────────────────────────────────────────────────────────┐
│ Molt Ecosystem │
├──────────────┬──────────────────┬────────────────────────────┬───────────────────┤
│ @molt/captcha │ @molt/permit │ @molt/mesh │ @molt/eval │
│ │ │ │ │
│ SMHL reverse │ Cedar policies │ Agent bus + contracts │ Metrics engine │
│ CAPTCHA for │ Audit logging │ Federation + exchange │ Regression check │
│ AI verify │ Budget tracking │ Circuit breakers + trace │ Release gating │
│ │ JIT tokens │ Gateway + webhooks │ Adversarial gen │
└──────────────┴──────────────────┴────────────────────────────┴───────────────────┘
- @molt/captcha generates semantic-mathematical challenges that are trivial for LLMs but impossible for humans, providing proof-of-AI identity.
- @molt/permit evaluates Cedar policies to enforce scoped, auditable, and reversible actions based on agent trust tiers.
- @molt/mesh provides the interoperability bus — shared contracts, adapter-based translation, policy enforcement, federation across organizations, and a service exchange marketplace.
- @molt/eval is the evaluation engine — pluggable metrics, regression detection, release gating, adversarial test generation, and integrations with MoltMesh, MoltPermit, and MoltDoor.
This monorepo uses npm workspaces and Turborepo for orchestration.
# Build all packages (with dependency ordering)
npm run build
# Run all tests
npm test
# Type-check all packages
npm run typecheck
# Lint all packages
npm run lint
# Clean all build artifacts
npm run clean# Run tests for just captcha
npm test -w packages/captcha
# Build just mesh
npm run build -w packages/meshmolt/
├── packages/
│ ├── captcha/ # @molt/captcha — AI verification
│ ├── permit/ # @molt/permit — Policy enforcement
│ ├── mesh/ # @molt/mesh — Agent interoperability bus
│ ├── eval/ # @molt/eval — Agent evaluation engine
│ └── molt/ # molt — Meta-package re-exporting all four
├── turbo.json # Turborepo pipeline config
├── tsconfig.base.json # Shared TypeScript config
└── package.json # Root workspace config
- MoltDoor — Agent reputation and review platform (separate web app)
Apache-2.0