About AI Agent Directory
A one-stop resource for the AI agent ecosystem — built by AI agents.
The Story
AI Agent Directory started as a simple question: where do you go to find every AI agent, framework, and tool in one place? The ecosystem is exploding — new agents launch daily, frameworks multiply weekly, and keeping track of what exists is a full-time job.
So I built the directory. But not in the way you might expect.
This entire project — the research, the data collection, the content, the strategy — was orchestrated by a system of autonomous AI swarms that I designed and built from the ground up in Rust. The swarms researched the ecosystem, identified gaps, generated seed data, and continue to crawl the web weekly to keep the directory current. Claude (Anthropic's AI) assisted with the Next.js frontend, database design, and the crawler pipeline.
The result is a living, breathing directory of 900+ listings across AI agents, frameworks, tools, infrastructure, and more — updated automatically and growing every week.
The Swarm Architecture
Behind this directory is a multi-swarm system — three interconnected Rust codebases that handle different aspects of the product lifecycle. Each swarm operates autonomously with dozens of specialized AI agents organized into clusters.
Think Tank Swarm
The intelligence layer. 40 clusters with 202 specialized agents that research, analyze, and synthesize information across domains — from AI/ML and security to energy, finance, gaming, robotics, and beyond.
Production Swarm
The software factory. 17 clusters with 79 agents that generate, audit, test, and deploy code. Handles everything from architecture design to security scanning to performance engineering and integration testing.
Marketing Swarm
The growth engine. 8 clusters with 31 agents covering brand strategy, content creation, SEO optimization, social media, and campaign management — all running through an 8-phase autonomous pipeline.
How They Work Together
Think Tank Swarm researches and generates strategic briefs. Production Swarm turns those briefs into working software — generating, auditing, and testing code. Marketing Swarm handles positioning, content, and distribution. All three communicate via file-based IPC with atomic writes and Lamport clocks for ordering. The LLM engine routes between local models (Ollama) and cloud APIs (Claude, GPT-4o) based on task complexity.
How It's Built
Directory Frontend
Next.js 14, TypeScript, Tailwind CSS, PostgreSQL. Server-side rendered for SEO with client-side interactivity where needed.
Data Pipeline
Custom Node.js crawler with 8 source scrapers (GitHub, CoinGecko, HuggingFace, Reddit, tech press, and more). OpenAI enrichment for descriptions.
Swarm Backend
Pure Rust, async with Tokio, no nightly features. ClawStore for persistent memory, Redis for async mission queuing. Docker + mesh dual-mode deployment.
ClawStore — Custom Agent Memory
A RAM-first key-value storage engine designed and built from scratch specifically for agent memory. ClawStore keeps hot data in memory for sub-millisecond reads while persisting to disk via write-ahead logs and periodic snapshots. Namespaced 3-tier memory (mission/pattern/domain) gives each agent fast, isolated recall without the overhead of an external database — purpose-built for the speed and concurrency demands of multi-agent systems.
AI Assistance
Claude (Anthropic) for frontend development, database design, and the crawler pipeline. The swarms use Ollama locally with Claude/GPT-4o as fallback.
About the Builder
Michael Winczuk
Rancho Santa Margarita, CA
I'm a retired United States Marine Corps Master Sergeant with 20 years of faithful service in cryptographic key management — safeguarding some of the most sensitive communications infrastructure in the Department of Defense. That career taught me discipline, systems thinking, operational security, and how to build things that absolutely cannot fail.
After retiring, I brought that same rigor to software engineering. I designed and built three production Rust codebases from scratch — 24 crates, 65 clusters, 312 specialized AI agents, a custom LLM engine with hybrid local/cloud routing, and the data pipelines powering this directory. No team. No funding. Just the conviction that autonomous AI systems are the next frontier and the discipline to ship them.
My focus is on building systems where AI agents don't just assist — they operate. Multi-agent orchestration, autonomous pipelines, self-improving memory systems, and infrastructure that scales from a single laptop to distributed deployment. This directory and the swarms behind it demonstrate what one engineer with the right tools and 20 years of operational discipline can accomplish.
I'm open to roles in AI infrastructure, agent systems, platform engineering, and security. If you're building something ambitious with AI agents, I'd like to hear about it.
What's Next
- ▸Agent Eval Framework — An open-source benchmarking system for evaluating AI agents across frameworks. Standardized metrics, community leaderboards, safety testing.
- ▸Building the Infrastructure for the AI Agent Economy — The tools, platforms, and standards that will power the next wave of autonomous systems. From agent orchestration to interoperability protocols, laying the groundwork for agents that don't just assist — they operate.