Fix Agent behavior

If your Agent responds or acts unexpectedly, use this table to find a likely cause. First reproduce the issue in a short test call.

Symptom	First check	Corrective action
Uses the wrong identity or tone	Persona field and {persona} in the prompt	Correct the Persona and check the shared Flow Persona if used.
Repeats a question	History limit and {conversation_history}	Keep recent turns and instruct the Agent to reuse confirmed facts.
Invents company information	Retrieved context and fallback rule	Connect the correct Knowledge Base and specify what to say when information is missing.
Claims an email was sent	Tool result or post-processing outcome	Confirm success only after the configured delivery succeeds.
Ignores a tool	Assignment, description and required inputs	Assign the tool and define when to use it.
Cuts off answers	Completion budget and voice events	Check model-specific requirements before increasing the budget.
Mishears names	Recognition language and vocabulary	Test specialist terms, spelling and caller corrections.
Interrupts callers	Turn detection and silence settings	Adjust one control and repeat a paused response.
Switches specialists unexpectedly	Routing result and previous intent	Review intent boundaries in the Flow.

Check changes with a new call

Save your change and start a new call. In Voice Debug Monitor, use Call Setup to check the settings used.

Test a failing tool in its own test panel first. Then check in a call that the Agent collects the required details and uses the tool at the right time.

Ask for help with a reproducible case

Provide the call identifier, approximate time, Agent or Flow, expected behavior and actual behavior. Share a minimal sanitized example, without credentials or unnecessary caller data. Continue with general troubleshooting for routing, uploads and delivery issues.