Getting started
Introduction
AgentSwarms is a learning and prototyping platform for agentic AI. You build real agents and multi-agent swarms, run them against real models, and study the traces — all in the browser.
The platform pairs a structured curriculum with a working lab. Every concept taught in the lessons maps to something you can open and run: a provisioned template, a swarm on the canvas, or a notebook. It is a learning and proof-of-concept environment, not a production runtime — the patterns transfer (everything exports to LangChain, LangGraph, CrewAI, Strands, and OpenAI Agents SDK code), the hosting does not.
Platform map
Each screen has a matching page in these docs:
| Surface | What it does |
|---|---|
| Dashboard | Workspace overview: quick actions, featured swarms, activity, recent runs. |
| Templates | 20 standalone agents and 30+ multi-agent swarms you can provision in one click. |
| Agent Builder | Configure agents: model, prompt, tools, knowledge, guardrails, memory, export. |
| Voice Agents | Talk to your agents — the speech-to-speech loop, the Voice tab, and production cloud stacks. |
| Swarm Canvas | Visual editor for multi-agent graphs — wire agents, routers, loops, approvals. |
| Notebooks | 97 runnable TypeScript notebooks covering frameworks and patterns. |
| Playground | Chat with any of your agents, with per-session overrides and trace links. |
| Traces & analytics | Every run recorded: model calls, tool calls, tokens, cost, latency. |
| Community | Publish your agents and swarms; remix what other learners ship. |
| Embeds | Ship any agent or swarm as an iframe widget on your own website. |
Your first ten minutes
- Sign in — free, no card. New accounts can run models through the built-in AgentSwarms AI provider immediately, with no API key.
- Provision a template from Templates — try Product Support Assistant (RAG with citations and an approval gate). It arrives with its knowledge base and tools pre-wired, plus a guided tour.
- Run it in the Playground and ask it something its knowledge base covers.
- Open the trace (Logs & traces) and read what actually happened: the resolved prompt, each tool call, tokens, and cost. Reading traces is the core skill this platform teaches.
- Then break it — edit the agent in the Agent Builder, or open a multi-agent template on the Swarm Canvas and change the graph.
How these docs are organized
Getting started covers the workspace itself. Build covers the surfaces where you create things — templates, agents, swarms, skills, notebooks. Run & observe covers executing and debugging. Platform covers community, certification, and provider integrations. Pages link to the live screen they describe, so the fastest way to read them is with the app open in another tab.