Skip to main content
When you test a draft in the Playground, ConvoTool shows you exactly how the agent produced each answer — which tools it called, what knowledge it pulled in, and what it cost. This makes it easy to debug responses that aren’t quite right.
The Playground resolved prompt drawer and trace details for a response

​
Trace chips

Each agent reply can show inline trace chips that summarize what happened on that turn:
  • Tool calls — which tools ran and what they returned
  • Routing — how the message was routed
  • Knowledge retrieval — that the agent pulled from your knowledge base
  • Escalation — that the conversation was flagged for a human
  • Errors and fallbacks — when something went wrong and how the agent recovered

​
Resolved prompt drawer

Click any agent reply to open the resolved prompt drawer. It has three tabs:
If an answer is wrong, check the Context tab first — it tells you whether the agent retrieved the right knowledge. If the knowledge is missing, add it from the Knowledge Base.
At the bottom of the chat panel, a cost footer shows the session’s token usage, cost in USD, number of tool calls, and number of routes — so you can see how configuration changes affect cost as you test.

Back to Playground

Return to the Playground overview.