Testing your agent before going live — and after every change — is the single most effective way to improve call quality. Test Mode lets you have a real voice conversation with your agent directly in the browser, with no phone required. This article covers how to use it and how to test effectively.

How Test Mode works

Go to Agents → Edit → Test Mode. Click Start Voice Call. Your browser will request microphone permission — allow it. You're now connected to your live agent, speaking to it exactly as a caller would.

voiceforge.pro/agents/agent_xxx — Test Mode
📞
Ready to talk?
Talk to your live AI voice agent right from the browser.
Uses your mic — please allow permission.
📞 Start Voice Call
ℹ️
Test calls appear in your dashboard
Every Test Mode call is logged as a real call in your dashboard and Analytics tab. This is useful for quality checking — you can review the full transcript and summary after the call. It also uses your plan's minutes, so don't run dozens of 10-minute tests.

What to test — a complete checklist

Don't just call and say "hello." Test Mode is most valuable when you deliberately push the agent into situations it will encounter with real callers.

Basic functionality

  • Does the agent answer with the correct greeting?
  • Does it introduce itself with the right name?
  • Does it mention your business name correctly?
  • Does it capture your name when you give it?
  • Does it ask for a callback number?

Knowledge tests

  • Ask about a service you definitely offer — does it answer correctly?
  • Ask about your prices — does it give the right answer or follow your policy?
  • Ask about your service area — does it correctly include and exclude locations?
  • Ask a FAQ you added — does the answer match what you wrote?
  • Ask about something from your scraped website — is it accurate?

Edge cases — things that will happen on real calls

  • Ask something outside your services — does it politely redirect rather than hallucinate an answer?
  • Ask for a price when you don't give prices — does it follow your system prompt's instruction?
  • Interrupt the agent mid-sentence — does it stop and listen, or keep talking over you?
  • Give an address outside your service area — does it tell you you're outside coverage?
  • Claim an emergency — does it follow your emergency protocol from the system prompt?
  • Be vague or unclear — does it ask a clarifying question rather than guessing?
  • Ask the same question twice — does the agent stay consistent?
  • Say nothing for 5 seconds — what happens? Does it prompt you?

Data capture

  • Provide your name mid-conversation — does the agent use it afterwards?
  • Complete a full call — does the dashboard summary show accurate data?
  • Is the urgency level correctly assigned in the call record?

What to do with what you find

After each test, open the call record in your dashboard and read the transcript. For every issue you found:

❓
Wrong or missing knowledge
→ Add or update a FAQ, re-scrape the website, or upload a document with the correct information. Then sync.
🗣️
Wrong call handling
→ Update the relevant section of your system prompt. Be more specific about what to do in that situation. Then sync and test again.
⏱️
Timing or interruption issues
→ Adjust silence timeout or interruption settings in Call Behaviour Settings.
🎤
Voice doesn't feel right
→ Go back to voice selection and try a different character. You can switch without losing any other settings.

Final check before going live

Before forwarding your real business number, run through this final checklist in Test Mode:

  • Complete a call from greeting to goodbye — does everything flow naturally?
  • The agent correctly captures name, number, and job type in one call
  • The call summary in the dashboard is accurate and useful
  • The opening greeting sounds natural when spoken aloud
  • You've tested at least one emergency or urgent call scenario
  • You're happy with the voice and would be proud for a customer to hear it
💡
Ask a colleague to test as a real customer
You know too much about your own agent to test it objectively. Ask someone who doesn't know what you've built to call it with a real question. Their experience is much closer to what your customers will have. Pay attention to any moment they feel confused or frustrated — that's your next improvement.

Testing is ongoing — not a one-time step

The best agents are tested regularly, not just at launch. Build a habit of testing after every system prompt change, after adding major new training material, and once a week during the first month of operation. After that, test whenever something in your business changes — new services, new prices, new policies.

→ Now that your agent is tested and live, see Your calls dashboard to understand all the data your agent is generating.