AgentSwarms

AgentSwarms is a coding tool that learns agentic AI by running, forking and breaking real agents. AgentSwarms is a hands-on browser platform for learning agentic AI, with 8 tracks, 50+ lessons, runnable agents and swarms, build-along labs, notebooks and failure-mode exercises.

AgentSwarms is a browser-based learning and prototyping platform for people who want to understand AI agents by working with them rather than watching videos. The curriculum spans 8 tracks and more than 50 lessons, and each concept is paired with a runnable agent or multi-agent swarm that opens directly in the browser. Topics include system prompts, tools and function calling, RAG, guardrails, ReAct and Reflexion patterns, Graph RAG, text-to-SQL and orchestration. The learning formats go beyond lessons. There are 21 animated presentation decks, 25 build-along labs that construct 10 standalone agents and 15 swarms in the real builder, and 97 runnable TypeScript notebooks covering LangChain, LlamaIndex, the OpenAI Agents SDK, the Vercel AI SDK and Google ADK. A set of Failure-Mode Labs hands over deliberately broken swarms, such as a looping tool call or a hallucinating retrieval step, and the platform checks whether the learner's repair works. Interactive blog posts add live calculators and simulators. In practice, a learner picks a template, such as an operations swarm, a market report team or a code-release pipeline, and gets a working system provisioned in seconds. A side-panel tour suggests prompts, then the learner forks it, changes prompts or models, attaches a knowledge base and watches traces in an observability dashboard. Agents can also be given a voice through a speech-to-speech loop, and any agent or swarm can be embedded on an external site through a domain-locked iframe. The site states plainly that it is a learning and proof-of-concept environment, not a production runtime. It suits developers, analysts and technical product people who want practical grounding before committing to a framework. Compared with video courses, it emphasizes hands-on experimentation, and compared with raw framework docs it supplies guided context. A self-hostable open-source platform for agents, RAG, evals and BI is also offered for teams that want to run it themselves.