Multi-Agent Lab
Five AI agents, running on your own GPU, build you a study plan through AI School. Give them a goal and watch the whole thing happen live: the Planner breaks the goal down, the Librarian calls a real search tool over the live course catalog, the Architect sequences the weeks, the Quiz Master writes your checkpoints, and the Reviewer fact-checks every course the others cited. No API key, no server, nothing uploaded.
1. Give the team a goal
2. The team at work
3. Message bus and tool calls
Every handoff between agents and every tool call shows up here, in order.
4. What the team produced
Your week-by-week plan, quiz checkpoints, and the Reviewer's grounding check will appear here.
Copied
What this demo is actually showing. This is the standard multi-agent pattern: one shared goal, specialist agents with narrow jobs, a message bus between them, and real tools that touch real data. The Librarian and the Reviewer are not language model calls at all, they are deterministic tools, because retrieval and verification are exactly the jobs you should never leave to a small model. The Reviewer's grounding check is the important part: it takes every course the Architect named and looks it up in the live AI School catalog, so a course the model invented gets flagged instead of shipped. That single step is the difference between a demo and something you would let a colleague use.
New to agents? The AI Agent Lab takes this apart across ten interactive pages: the agent loop, tool calling, memory, planning, routing, testing, and guardrails. For the written course, start with the AI Agents track, then Agent Frameworks in Practice. To see the raw model on its own, open the Browser SLM Lab.
New to agents? The AI Agent Lab takes this apart across ten interactive pages: the agent loop, tool calling, memory, planning, routing, testing, and guardrails. For the written course, start with the AI Agents track, then Agent Frameworks in Practice. To see the raw model on its own, open the Browser SLM Lab.