Agent playground

The Agent playground (at /agents) is where you run and interact with an agent on its own — outside of any graph. It is the quickest way to try out an agent definition, iterate on its behavior, and watch how its memory evolves as you talk to it.

Opening agents in tabs

The playground has an AgentSidebar listing your agent definitions. Open one and it appears as an AgentTab in the main area, so you can keep several agents open at once and switch between them. Each tab is an independent Playground session for that agent.

Agents overview How agent definitions are configured and what they can do.

Running an interaction

1. Pick an agent from the AgentSidebar to open it in a tab.

2. In the AgentPlaygroundArea, send a message to the agent and read its response.

3. Continue the conversation — the agent runs with its configured model, tools, and memory, just as it would inside a graph.

Inspecting memory and state

The AgentMemoryInspector shows the agent’s memory and state as you interact. Use it to see what the agent is retaining across turns, verify its memory strategy is behaving as expected, and debug cases where it forgets or over-remembers context.

Memory strategies How agents remember across turns, and how strategies resolve.

Tip

The playground is the fastest feedback loop for tuning an agent before you drop it into a graph or bind it to a chat. Run it here, watch the memory inspector, then refine the agent definition.