Agent Playground Overview
Where to start: what Agent Playground is, and the guides for creating, building, and running an agent.
What is Agent Playground?
Agent Playground is the visual builder under Agents in the sidebar, where you assemble a multi-step agent from nodes on a canvas, run it, and read what each step produced. Reach for it once a single prompt can’t do the whole job on its own, such as when one step’s output needs to feed the next step, or a step needs to call another agent.
The graph you build on the canvas is that agent, one step per node. Every node is one of two types:
- LLM Prompt runs a prompt you already built in Prompt Management
- Agent Node runs another saved agent as a single step
Start here
Create an agent
Create your first agent and manage its versions
Build a workflow
Add nodes, configure them, and connect them on the canvas
Run an agent
Run an agent and inspect what each step produced
Understanding Agent Playground
How graphs, nodes, ports, and edges fit together
Versions & Execution
How the draft/active version lifecycle and execution model work
Something not working, or need to know a limit before you build? See Agent Playground FAQ & fixes and Limits & rules.
Questions & Discussion