For teams shipping AI agents

Your agent's next thousand conversations, before your customers have them.

Simulated users who are confused, impatient, off-topic or adversarial, in the mix you actually see in production. Find the failures in a test run, not a support ticket.

Join the waitlist

How it works

  1. Describe who talks to your agent

    Start from a description, or ground the simulated users in your real transcripts so they ask what your customers ask, the way they ask it.

  2. Run conversations at scale

    Hundreds of multi-turn conversations in parallel, in CI or on demand, with every transcript kept.

  3. See where it breaks, and how real the test was

    Failures grouped by type, plus a check of how closely the simulated users resemble your real ones.

Why scripted tests miss what users do

Hand-written test conversations are polite, on-topic and short. Real users change their minds halfway through, paste in error messages, ask two things at once and give up. An agent that passes a script can still fail the people it was built for.

  • Coverage: test the long tail of personas and intents, not the ten you thought of.
  • Regression: rerun the same simulated population after every prompt or model change.
  • Realism you can measure: we compare simulated conversations with your real ones, so you know how far to trust a test.

Join the waitlist

We're building this now. Tell us what you'd simulate and we'll reach out as early access opens. The most interesting use cases get in first.

We'll only use your email to contact you about early access. Privacy.