Skip to main content

​
Test Call

The second tab of the right panel. It’s a text chat, not a voice call: you type as if you were the caller and walk the flow.
📷 Screenshot needed — the Test Call tab with a test conversation and a tool execution.Save to /images/en/voice-agents/test-panel.png
Attached tools run for real. This is not a simulation:
  • A calendar or integration tool that writes will create, move or cancel a real appointment.
  • A transfer dials a real number.
  • Send WhatsApp Template sends a real WhatsApp message.
  • End Call ends the test session.
Use test data, or temporarily detach the tools that write.

​
When you can test

The panel requires three things:
The agent must be created — you can’t test before saving it the first time.
It must be synced with Aurora.
There must be no unsaved changes. The test runs against the saved configuration, not what’s on screen.
That last one trips people up most: if you edit the prompt and test without saving, you’d be testing the previous version. That’s why the panel blocks and asks you to save first.

​
Cost and limits

It’s free. The test panel consumes no credits — it’s a text conversation, not voice minutes.
It’s capped at 60 turns per hour per user. When you hit it you’ll see “Test chat limit reached for this hour”.
Testing requires write permissions on agents, precisely because tools run for real and can create records in your connected systems.

​
What you’ll see

Tool calls appear as expandable cards showing the arguments the agent used and the result. It’s the fastest way to see why the agent chose to use — or not use — a tool. If the agent hangs up, the conversation ends and you have to Reset to test again.

​
Summary

The third tab. A checklist of what’s still missing — name, description, model, language, voice and prompt — and a recap of the configuration: per-minute price, assigned number, direction, transcription mode, voicemail detection, tools and tags. Useful as a final check before activating.

​
The per-minute price

In the top bar, next to the name. It recalculates as you edit and covers the model, the voice and the length of the prompt.
The prompt affects the price. A very long prompt can exceed the model’s token limit, and beyond that the per-minute rate scales upward. The chip’s token bar fills up, but no explicit warning appears.If you see the bar maxed out, shorten the prompt or the FAQs.
FAQs count towards this calculation, because they’re sent alongside the prompt.

​
Activating the agent

Saving publishes immediately. There’s no server-side draft and no version history: when you press Update Agent, the agent handling live calls changes at that moment.
The only lifecycle control is the Active / Paused toggle in the top bar.
For big changes to a live agent, duplicate it, work on the copy, test it, then swap which one is active.
When you duplicate an agent, the copy loses part of its provider-side configuration — the prompt and the voice settings — until you save it once. Always open the copy and press save before using it.

​
Import and export

From the agent list you can export an agent to JSON (choosing which tools to include) and import it later. Importing an Aurora agent makes the phone number mandatory, and an inbound one also needs a schedule.

​
What doesn’t exist

So you don’t go looking:
  • Version history or rolling back to an earlier configuration.
  • Knowledge base or document upload.
  • Call recording, maximum duration or retries as agent settings.
  • Analytics inside the agent — per-agent metrics live under Analytics, filtered by that agent.

​
Back to the start

Introduction to voice agents

The full path through this section.