Connection Types
How all four connection types compare — phone, LLM Chat, Web Chat, Web Voice.
Testzilla tests agents on a real phone line in two directions. The direction is always named from your agent’s (the system under test) point of view:
These two are not the same mechanism with the dial direction flipped — they work differently end to end. Inbound is driven live by the tester; outbound is evaluated after the fact from a post-call webhook. The comparison below makes the practical difference clear so you can pick the right one.
| Inbound Call | Outbound Call | |
|---|---|---|
| Availability | Available today | Coming soon |
| Who places the call | Testzilla dials your agent | Your agent dials a Testzilla number |
| Who answers | Your agent (the SUT) | The test’s predeployed Testzilla agent |
| How the conversation is driven | The TZ tester drives it live, in real time | The predeployed agent runs the call autonomously |
| When evaluation happens | During / right after the live call | After the call, from a post-call webhook transcript |
| Phone number | You provide your agent’s number | Testzilla provisions a dedicated number per test |
| What you configure | Provider + your agent’s number + credentials | Outbound test (the dedicated number is auto-provisioned) |
| Per-call billing | 40 credits per call (the phone rate) | Same per-call phone rate as inbound |
| Subscription | None beyond your plan | $5/mo per outbound test (dedicated test-line add-on) |
In short: choose Inbound when your agent answers calls and you want the tester to drive a live conversation. Choose Outbound (once available) when your agent is the one that initiates calls and you need to verify what it does when it places a call.
Your agent receives the call. Testzilla dials in to your agent’s phone number through your telephony provider (Retell or VAPI), and the AI tester drives the conversation in real time — speaking, listening, and steering toward the scenario you wrote — then evaluates and scores the result.
This is the highest-fidelity path for an agent deployed on a number that real callers ring. The whole voice pipeline — telephony, ASR, LLM, TTS — is exercised exactly as a real caller would experience it, including IVR routing, hold music, and transfers.
How it works
When to use it
What you configure
Billing
See the Test a Phone Number guide for the full step-by-step.
Your agent places the call. Instead of Testzilla dialing your agent, your agent dials a phone number that Testzilla provides, and a predeployed Testzilla agent answers and runs the conversation. Because nobody is driving the call live from Testzilla’s side, evaluation happens after the call from a post-call webhook.
This is genuinely different from inbound, in three ways worth understanding before you plan around it.
Each outbound Test gets its own dedicated phone number:
So an outbound test is not just a scenario — it is a scenario bound to a live phone line that exists for as long as the test does. That dedicated line is what your agent dials, and it is also why outbound carries a small monthly subscription (see Billing below).
When your agent (the SUT) originates the call to the dedicated number:
After the call completes:
Because scoring runs on the post-call transcript rather than a live session, the result lands after the call ends rather than mid-conversation.
When to use it (once available)
What you configure
Billing
Outbound has two billing components:
The fastest way to think about it:
Match the direction to how your agent actually behaves in production: test inbound if real users call your agent, test outbound if your agent calls users.
Connection Types
How all four connection types compare — phone, LLM Chat, Web Chat, Web Voice.
Test a Phone Number
The full step-by-step for inbound phone testing.
Creating Channels
Set up a Phone channel for your agent.
Pricing
Plans and credit costs per connection type.