agentswarms.fyi

AgentSwarms is an interactive learning platform for agentic AI, built around runnable agents and swarms. Explore a curriculum of 50+ lessons paired with 50+ ready-to-run examples, including ReAct, Reflexion, Graph RAG, and text-to-SQL.

Instead of only watching tutorials, you open templates in your browser and immediately practice by reading the lesson side-panel, forking the agent or swarm, and experimenting. You can also diagnose and repair deliberately broken swarms in Failure-Mode Labs to learn common failure patterns like hallucinating RAG, runaway loops, dead branches, and more.

It also includes a visual Agent Builder / drag-and-drop swarm canvas with guardrails, tools, memory, RAG, evals, and human approval workflows. Learn by building and breaking in a non-production learning and POC environment—no installs or API keys required to start.

Do I need to install anything or use API keys to start?

No—start by opening live demo or curriculum templates in your browser, with guided prompts and no API keys to begin.

What can I learn with AgentSwarms?

You can learn core agent concepts and patterns while running runnable examples, including prompts/system messages, tools & function calling, Graph RAG, text-to-SQL, and multi-agent swarms.

Are the lessons interactive or just videos?

Lessons are interactive: each concept is paired with a runnable agent or swarm you can open, run, and fork right away.

What are Failure-Mode Labs?

Failure-Mode Labs are guided practice where you diagnose and repair deliberately broken swarms, learning how issues like hallucinated RAG and runaway loops happen and how to fix them.

Can I modify templates and experiment?

Yes—fork templates, tweak settings like prompts, and break things to observe how changes affect behavior using the platform’s tracing and observability.

Is AgentSwarms for production deployment?

No—it's a learning and POC platform focused on helping you build, run, and understand agentic systems rather than acting as a production runtime.