H

hitl

paper2code is an AI agent skill that converts arXiv papers into citation-anchored, runnable code. It includes ambiguity auditing, appendix mining, and human-in-the-loop enhancements contributed by Roxanne Ardary for interactive code refinement, experiment configuration, and verification of all implementation decisions.

Page-Agent is a JavaScript library that transforms any web page into an AI-driven interactive agent. It allows users and developers to control page elements and workflows using natural language commands, powered by large language models. With features like DOM manipulation, client-side execution, and optional human-in-the-loop confirmations, Page-Agent simplifies automation and enhances accessibility. The latest update includes end-to-end encryption (E2EE) for secure, private interactions.

Covenant AI is an open-source artificial intelligence infrastructure platform focused on human authority, bounded autonomy, transparency, privacy, and accountable AI execution. Built as a complete alternative to traditional AI agent frameworks, Covenant AI ensures that every AI action is authorized, explainable, revocable, and verifiable — creating intelligence systems that remain fully accountable to the people they serve.

Open Arsenal provides open, self-hosted software specifications designed for deployment across private and public networks. Built for portability, interoperability, and transparency, each specification gives organizations the freedom to implement and customize solutions without vendor lock-in while remaining free to use under the AGPL-3.0+ license with required attribution. Specification Branding Licenses available upon request.

Vexa is a privacy-first, human-in-the-loop AI orchestration platform that connects and automates enterprise tools through a secure, auditable control layer. It enables users to build and execute multi-step workflows across systems like Slack, Jira, and Salesforce while enforcing strict privacy controls, zero-trust data handling, and full execution transparency through explainable, replayable actions.

CortexLoop is a human-in-the-loop AI orchestration system designed to keep all reasoning, planning, and execution fully transparent, structured, and under human control. It transforms AI from an autonomous or opaque assistant into a governed cognitive loop where multiple solution paths are generated, evaluated for risk and confidence, and returned to the user for decision-making. Every action is logged through a decision ledger, enforced by a policy engine, and constrained by strict system rules including AGPL+ licensing and attribution requirements. The result is a traceable, auditable AI system that prioritizes clarity, accountability, and human authority at every step.

Atlas Continuum System is a spec-driven operating system for building persistent narrative, visual, and physics-based worlds. Instead of relying on prompt-to-output generation, it compiles structured narrative specifications into consistent simulations using identity-locked character agents, reusable visual assets, and law-governed physics. Every change is traceable, governed through human approval, and designed to preserve continuity across stories, worlds, and time.

Autonomous Loop is a self-hosted Continuous Improvement Operating System designed for air-gapped AI factories. It enables organizations to continuously monitor, evaluate, govern, and improve AI applications through observability, data drift detection, human review workflows, autonomous improvement agents, and intelligence lifecycle management while keeping models, data, and operations under full control.

AuthenticAI is an open-source, locally hosted AI governance platform designed to keep intelligence local and decisions accountable. It provides a trust layer between AI systems, autonomous agents, enterprise data, and external tools through metadata filtering, data minimization, retrieval trust scoring, model routing, human approval workflows, and MCP governance. AuthenticAI enables organizations to deploy AI securely with transparency, control, and accountability.

Human Intent Governance Standard (HIGS) is a structured framework for building AI systems that operate under verified human intent, enforceable policy, and auditable compliance. It ensures that every decision passes through intent clarification, legal and policy validation, risk assessment, and human approval before execution. Designed for regulated and high-trust environments, HIGS prioritizes transparency, explainability, and human accountability while enabling modular, extensible governance across legal, technical, and organizational domains.

Human Decision Specification (HDS) is an open, modular standard for building AI systems that keep humans in control of decision-making. It defines a structured workflow where AI systems ask clarifying questions, discover and confirm human intent, evaluate legal and compliance constraints, and provide transparent recommendations. HDS ensures that every significant action is approved by a human, fully auditable, and designed to prevent autonomous decision-making without oversight.

PHITrack is an open-source platform that monitors, analyzes, and visualizes PHI (Protected Health Information) transmissions using AI-driven insights while maintaining human-in-the-loop oversight. It provides dashboards, alerts, predictive risk scoring, and synthetic dataset simulations to ensure HIPAA-safe compliance, training, and reporting.

Nexalis is a specification intelligence system that transforms natural language intent into structured, modular, and time-aware system designs. It enables human-guided creation of evolving specifications that are simulated, stress-tested against uncertainty, and continuously refined through adversarial analysis, emergence detection, and meta-learning feedback loops.

ListingStream is a modular, AGPL-3.0+ licensed visual commerce platform that uses AI to intelligently insert sponsored products into photographs while preserving the integrity of the original image. Built around mandatory human-in-the-loop approval, real-time legal compliance, explainable AI, immutable audit trails, and pixel-preserving rendering, ListingStream enables trusted, revenue-generating visual marketing for real estate and other industries through a flexible, compliance-first architecture.

Open-source LiveKit-based voice AI agent for outbound + inbound phone calls — native Indian-language coverage via Sarvam. MIT, part of Mesh Pilot.

Open-source AI agent for multi-platform social posting + ORM (YouTube Shorts, IG Reels, X, LinkedIn). HITL approval gate. MIT, part of Mesh Pilot.

Open-source AI agent for autonomous Shopify SEO — audits, schema, content gaps. Ships as a Shopify Embedded App. MIT, part of Mesh Pilot.

Open-source AI agent for outbound B2B sales — discovery, enrichment, LLM drafting, HITL approval, send. MIT, part of Mesh Pilot.

Open-source AI agent for paid-acquisition automation across Meta, Google, TikTok, Amazon, LinkedIn. HITL approval gate. MIT, part of Mesh Pilot.