AgentKits
Resources

Building & Governing AI Agents

A working map for taking an AI agent from idea to something safe enough to run in production. It mixes what we build — the open AgentAz standard and free tools — with the external references that actually matter, so it's a starting point you can trust rather than a wall of our own links.

The short answer

An AI agent's safety is a function of its autonomy: what it can do without a human. The AgentAz Trust Levels name that on a scale — ADV (advisory, can't change anything), A0, A3 (every consequential action human-approved), A4 (bounded autonomy), A5 (full autonomy). You declare it in an agentaz.json, verify it with the scanner, and enforce it in CI with the GitHub Action. It's framework-independent — the same file describes an agent built in LangGraph, CrewAI, or n8n. A standard, not a runtime.

Starter files (download & drop into your repo)

Adopting the standard is a one-file step. Grab a template, fill it in, and — if you want enforcement — add the CI workflow. These carry the governance vocabulary into your own project, not just ours.

How do you classify an AI agent’s autonomy?

AgentAz is an open specification for declaring an agent's authority — its worst-case action, Trust Level (ADV→A5), approval gates, and tool boundaries — so a team, or an LLM, can reason about its risk before deploying.

Free tools

Reference blueprints

Sixty-two governed agent blueprints across thirty categories, each with a declared Trust Level and worst-case action. Useful as starting points or as examples of what a well-specified agent looks like.

Guides

What standards and frameworks apply to AI agents?

The references enterprises and security teams actually use. Worth knowing regardless of which standard you adopt.

External canon: agent frameworks

  • LangGraph — Graph-based agent orchestration with first-class interrupts for human-in-the-loop.
  • CrewAI — Role-based multi-agent orchestration.
  • Model Context Protocol (MCP) — The open standard for connecting agents to tools and data sources.
  • OpenAI Agents — OpenAI's agent tooling and patterns.

Missing something you rely on? Suggest a resource — this map improves with real use.

From the Artifipedia network

AgentKits is part of Artifipedia, a sourced encyclopedia of AI concepts, models, and terms. If you are trying to understand the wider AI landscape behind agents — the models, techniques, and vocabulary — it is a good place to start.