How RingTru tests an AI voice agent

Simulated callers ring your agent and AI judges score every call. You get a short list of what to fix.

Get a free report Book a call

AgentsSample data

Test calls this month1,068 18% on last month

Pass rate93.3% 2.1 points on last month

Wrong callers let through5 4 fewer than last month

Median reply time0.98 s 0.06 s slower than last month

Claims line

Needs workRetell, inbound

test calls: 240
passed: 87.1%
median reply: 0.88 s

Reads policy numbers back wrong after a correction.

Booking line

PassingVapi, inbound

test calls: 506
passed: 96.8%
median reply: 0.94 s

Holds the new day when a caller changes their mind.

Billing line

WatchLiveKit, inbound

test calls: 322
passed: 92.5%
median reply: 1.21 s

Slows past two seconds when callers cut in.

Sample data. An early design of the RingTru dashboard.

Simulated callers

Accents, noisy lines, and callers who cut in or change their mind.

AI judges

Each call is scored against your written rules, with a pass or fail for each.

Evidence report

Failed calls are grouped by cause and ranked by risk, each with a fix.

Connect your agent

Point us at the agent and we ring it.

You give usYour agent IDAnd an API key for Vapi or Retell, or a phone number you may test.

We run20 to 2,000 callsTen at a time, with the callers you pick.

You getFindingsRanked by risk, each with quotes and a fix.

Run it on every release

Start runs from the dashboard, or from the automated checks that run before each release (CI). A change that breaks a behaviour stops the release.

See Pro and Scale plans

$ ringtru run --suite first-notice --fail-on highCalling the test number, 300 calls, 10 at a time2 behaviours below the pass mark3 findings, 1 high. Release check failed.

Early design. Commands may change before launch.

Run a test on your own agent.