About AgentKits
AgentKits is a free, open registry of production-ready AI Agent Blueprints for building real agents — and a practical resource for understanding and working with agents in the real world. You browse, read, copy, and download. No login, no paywall.
What a kit is
Every kit is a complete, readable spec plus a downloadable bundle. The page gives you the system prompt, architecture, the tools the agent needs, the run-loop workflow, implementation notes, and variations. The bundle packages all of it as files you can drop straight into a project: system-prompt.md, tools.json, workflow.md, .env.example, kit.json, a starter, and a README.
How to use it
Find a kit that matches your task across 60 kits in 30 categories. Copy the system prompt, copy the whole thing as Markdown, or download the zip. Then adapt it to your stack — Claude, OpenAI, LangGraph, n8n, CrewAI, or the Vercel AI SDK. Nothing here is locked to one framework.
Authoring with Bolt
Kits are plain JSON files, one per kit. Adding a kit is dropping a new file into the registry — no database, no build configuration. The repository ships a JSON Schema and a copy-paste authoring prompt so you can generate a valid kit with Bolt, validate it, and open a pull request. See the submit page to get started.
Why it exists
AI agents are becoming a core part of modern software and automation, but the space is still fragmented. Information is often too technical, too scattered, or not clearly tied to real-world use cases. AgentKits exists to reduce that friction — organizing and simplifying what actually matters so builders, founders, and teams can move from learning AI to applying it.
Who it's for
- People building AI-powered products.
- Developers exploring agent frameworks.
- Founders and startups working on automation.
- Teams looking to improve workflows with AI.
- Anyone trying to move from understanding agents to shipping them.
The blog
Alongside the registry, we publish hands-on writing on building agents — framework comparisons, build guides, tooling reviews, and the patterns that hold up in production. It's clear, practical, and neutral, without hype or overpromising. Read the AgentKits blog.
Governance built in: AgentAz™
Every blueprint is held to a defensive standard we document with AgentAz, a lightweight, design-time vocabulary that records what an agent is authorized to do — and why — so a person can review it before it ever runs. It is a vocabulary, not a framework: it pairs with whatever policy engine you use and enforces nothing itself. Every blueprint ships a reviewable agentaz.json boundary contract. See the AgentAz specification or try the Trust Level calculator.
Open source
AgentAz™ is fully open source. The reference implementation and JSON Schema are licensed under the Apache License 2.0 (chosen for its explicit patent grant, which makes it safe for companies and foundations to adopt), and the specification text under CC‑BY‑4.0. Our aim is for AgentKits to be the central, high-quality open registry for production AI agent governance — a neutral standard any team or vendor can build on. The specification, frozen v1.0.0 schema, changelog, and governance live on GitHub. AgentAz™ remains a trademark of AgentKits.
Principles
- Readable first. A kit you cannot understand is a kit you cannot trust.
- Framework-neutral. Patterns over lock-in.
- Human-in-the-loop by default, especially for regulated work.
- Static and fast. No accounts, minimal tracking, open to adapt.