Skip to main content
The playground lets you test your agent in simulated conversations before any real lead ever talks to it. You chat with the agent exactly as a lead would, watch how it responds, and use what you learn to refine your agent settings.

Opening the playground

Open the playground from your agent’s dashboard. Type messages as if you were a lead, and the agent replies using its current configuration: its goal, identity, business rules, variables, and knowledge base.

Conversations are simulated

Everything in the playground is a simulation. No real lead is created, no messages are sent to real contacts, and nothing you type reaches anyone outside the test.
The playground does not affect live leads. Chatting here never creates a lead, never sends a message to a real contact, and does not change any conversation already in progress.

Apply edits before testing them

The agent in the playground uses the prompt from the last completed training. After you change your Business rules in settings, click Train your agent and wait for training to finish. Until then, the playground still reflects the previous version of your rules, not your latest edits.

Iterating on your agent

Use the playground to find gaps, then fix them in settings and test again.
1

Play the lead

Start a conversation as Tony Stark would: ask the questions a real lead asks, and let Jarvis respond.
2

Test edge cases

Push on the hard parts: raise objections, go off-topic, and try the scenarios you are unsure about, so you see how Jarvis handles them.
3

Refine the right section

When a response is off, open the matching part of your Business rules in settings, for example Objection handling for pushback or Agent limits for things Jarvis should never say, and adjust the wording.
4

Train and re-test

Click Train your agent, wait for training to finish, then run the same conversation again to confirm the change worked.
Start with Business context, train, and test here, then refine the other sections of your business rules based on how Jarvis actually responds.
Last modified on June 17, 2026