Test a voice agent before going live | FieldCamp

Q: How do I test an AI voice agent?

Open the agent in the FieldCamp Voice builder and click Test. Use Test Chat to type to it, Test Audio to talk to it in your browser, Simulation to run saved caller scenarios, or Call my phone… in the ⋯ menu to get a real phone call.

Q: Which test method should I use first?

Start with Test Chat, because you can see each step, transition and tool call as you type. Move to Test Audio and a phone call once the conversation flows the way you want.

Compare the ways to test a FieldCamp voice agent before real callers hear it: browser audio, text chat, simulated test cases, and a call to your phone.

Before a voice agent answers real customers, test it the way a caller would reach it. FieldCamp Voice gives you four ways to do that, from typing a quick message in the builder to getting a real call on your phone.

All four run the agent's saved draft, not the published version. That means you can test changes safely before you publish them. Check that the builder header shows All changes saved before you start a test, so the test runs your latest edits.

Ways to test

Open the agent in the builder and click Test. The test panel has three tabs: Test Audio, Test Chat and Simulation. The phone test is in the builder's ⋯ menu as Call my phone….

Method	Where to find it	Best for	Shows in call history
Test Audio	Test → Test Audio	Hearing the voice, timing and interruptions in your browser	Yes
Test Chat	Test → Test Chat → Manual	Checking prompts, step transitions and tool calls quickly by typing	Yes
Simulated tests	Test → Simulation (or Test Chat → Simulated)	Rerunning the same caller scenarios after every change and getting a pass or fail per criterion	No
Phone call	⋯ → Call my phone…	Hearing the agent on a real phone line, as a caller would	Yes

A few differences matter when you choose:

Test Audio and phone tests are billed like live calls. Simulated tests aren't billed as call minutes, although they still call the language model.
Browser, chat and phone tests appear in the agent's call history alongside real calls, with no test label. Keep a note of when you tested so you can tell them apart.
In Test Audio, a transfer is faked: no number is called, and the agent carries on along the transfer's failure path. Test Chat refuses transfers. In a simulated test, a transfer needs a tool mock.
Other tools act for real during a browser test. If a step can create a request in FieldCamp, book a job or call your own system, the test does that too. Use test customers and data you can delete.
Simulated tests need an OpenAI text model.

Recommended order

Test Chat first. Typing is the fastest way to see which step the agent is on, which transition fired and which tool it called. Fix the wording and the flow here. See Test your voice agent in the browser.
Test Audio next. Once the conversation is right, listen to it. Check the voice, how long the agent waits before it replies, and what happens when you interrupt it.
Save simulated test cases. Turn the calls that matter most into repeatable cases, such as a no-heat emergency, a routine maintenance booking and a caller who won't give an address. Rerun them after every prompt change. See Run simulated tests.
Finish with a phone call. Hear the agent on a real phone before you publish. See Test your voice agent with a phone call.

Pre-launch checklist

Work through this list before you publish and send real calls to the agent, either from a number connected in Voice (see the quickstart) or, if you've connected FieldCamp, from a call flow.

Placeholders are filled in. Templates contain placeholders such as [Business Name]. If the agent says "This is [Business Name]" in a test, the placeholder is still in the prompt.
Your emergency rules are your own. What counts as an emergency, such as no heat in winter, an active leak or a burning smell from a panel, and who handles one, is your decision. Check the agent follows the rules you wrote.
Urgent and routine calls both work. Test at least one call of each kind, plus a caller who is vague or changes their mind.
Every step has the tools it needs. A step can only use the tools selected on that step. If the agent says it will do something and nothing happens, check the step's tools. See Add tools to your voice agent.
Variables have fallbacks. A variable the agent can't fill becomes empty on a real call. Values you enter under Settings → Variables are test values only and don't apply to real calls. Add a fallback, such as {{customer_name | there}}. See Use dynamic variables.
Transfers go to the right place. Browser tests fake transfers and follow the failure path, which lets you check what the agent does when a transfer fails. Confirm the destination numbers with your team. Warm transfer works only with Twilio. See Transfer a call to a person.
The agent doesn't promise what you can't deliver. Listen for arrival times, prices or callbacks you haven't approved.
Test records are cleaned up. Delete any requests or jobs your tests created.

When the agent passes, click Publish. For the full path from a new agent to a live call, follow the quickstart.

Frequently asked questions

How do I test an AI voice agent?

Open the agent in the FieldCamp Voice builder and click Test. Use Test Chat to type to it, Test Audio to talk to it in your browser, Simulation to run saved caller scenarios, or Call my phone… in the ⋯ menu to get a real phone call.

Which test method should I use first?

Start with Test Chat, because you can see each step, transition and tool call as you type. Move to Test Audio and a phone call once the conversation flows the way you want.

Do tests use my real tools and data?

Browser tests call your tools for real, apart from transfers, which are faked, so a test can create real records, such as requests in FieldCamp if you've connected it. Simulated tests use tool mocks instead. Use test customers and data you can delete.

Do test calls count toward usage?

Test Audio and phone tests are billed like live calls. Simulated tests aren't billed as call minutes, but they still call the language model.

Do tests run the draft or the published version?

Every test type runs the agent's saved draft. You can test changes before you publish them, and the published version keeps taking real calls while you do.

Test a voice agent before going live | FieldCamp

On this page