
The playground: a simulated conversation with your agent.
Open it
Open the playground from the agent’s page. The agent replies using its current configuration: goal, identity, business rules, variables, pipeline, and knowledge base.Everything here is simulated. No lead is created, no message reaches a real contact, and live conversations are not affected.
Run your real scenarios
Test the cases you wrote in your worksheet, plus the ones every agent must survive:Propose a change and see it applied
The most useful minute in any onboarding is when the customer says “I would like it to do X instead,” and watches the configuration change. Do it yourself:1
Spot the behavior you want to change
For example, the agent asked for the budget before explaining what the product does.
2
Ask NexorGPT for the change
Open the NexorGPT panel or page and describe it: “Explain the product in one line before asking for budget.” Confirm the before/after card.
3
Train the agent
Click Train your agent and wait for training to finish. The playground uses the prompt from the last completed training, so an untrained edit is invisible here.
4
Run the same conversation again
Repeat the exact scenario and confirm the new behavior. If it is still off, refine the wording of the matching business rules section rather than adding a new rule elsewhere.
Test on WhatsApp before connecting a number
Eligible accounts can open Test my agent, scan a QR code, and talk to the agent from their own WhatsApp. The session stays isolated and does not contact real leads. Use it to feel the timing and formatting of real messages.
Test my agent: talk to the agent from WhatsApp without connecting a business number.
Test connected agents
If you have built several agents that transfer leads between them, start the conversation with the first agent and drive it toward the transfer condition. Confirm that the next agent picks up with the conversation history and the fields you chose to copy. See Connected agents.What “done” looks like
You are ready to move on when:- every scenario in the table above behaves the way you expect;
- you have made at least one change from NexorGPT and seen it take effect after training; and
- you can tell, for any wrong answer, which block of the agent you would edit.