Overview
Agent testing & QA is Voiceflow’s automated testing product for checking customer journeys before release. Replay specific interactions with Tests or explore different user replies with AI-driven Simulations. A saved conversation can combine both. Check responses, routing and tool calls to find failures and catch regressions after changes.
Open Tests from your project sidebar, then New test to author either approach in the shared editor. Runs use the selected Environment; check it before interpreting a result.
Choose a starting point
- Reproduce a known interaction? Write scripted turns and add the response, routing or tool-call checks that define success.
- Explore different user replies? Define a Simulation scenario and success criteria, then set its conversation turn limit.
- Try the same behavior with different customer context? Choose a starting event and persona or starting variables.
- Check several saved conversations after a change? Run a batch and inspect the failed expectations.
Compare Tests and Simulations when deciding which approach fits a behavior.
Explore Agent testing & QA
Make important customer interactions repeatable. Script messages and check responses, routing, tool calls and execution order.
Explore how an agent handles different user replies. Define the simulated user, success criteria and conversation turn limit.
Check behavior for different customers and entry points using personas, starting variables and events.
Find which checks failed and why. Run one conversation or a batch, inspect outcomes and logs, and reopen or retry past runs.
Continue the testing workflow
- Investigate a failure: compare the expected behavior with the conversation, check reasoning and logs before changing the agent or its checks.
- Correct and rerun: edit the relevant behavior in Agent studio, then retry the affected conversations in the intended Environment.
- Extend coverage from production: use Transcripts to find a real case. Atlas can turn a conversation into a Test for you to review and run.
- Automate testing: use the QA API reference for developer-driven execution.
Runs consume credits. See Credits and costs before running broader coverage.