Documentation / Testing your agent
Testing your agent
Check responses, tools and flow using voice or text before publishing.
Test situations the agent should solve and situations that require clarification, refusal or handoff.
Prepare your test
Save pending changes and resolve validation errors shown in the editor. Open Test agent and choose voice or text.
Test with text
Send one question at a time and inspect the replies. Text helps investigate instructions and tools independently of audio. It does not check pronunciation, speech recognition or interruptions.
In manual text testing, use the rewind option when available to explore a different reply from an earlier point. End the session when finished.
Test with voice
Start the conversation, speak naturally and watch the transcript. Check names and numbers, whether the agent waits for you to finish, and whether interruptions follow your configuration.
Use a scenario checklist
| Scenario | Expected result |
|---|---|
| Valid order | The tool looks up the confirmed number and the agent explains the response |
| Missing number | The agent asks for it before looking up the order |
| Unavailable API | The agent explains the failure without inventing a result |
| Unrelated question | The agent explains its scope and the next step |
| Corrected customer detail | The action uses the corrected value |
Test tools may access real systems. Use test endpoints and credentials for actions that create records or send messages.
Review and publish
Open the run to check transcript, actions and available data. Record the failed question, change one instruction and repeat. Review completed conversations against the agent's goal; use manual tests to check behavior before publishing.
Continue with Publishing. The automatic simulator marked as under development is not required for this agent.