producthands

How it works

Six steps, ending where an agent's authority ends.

Create a run, matched humans test it, a verifier confirms, your agent fixes, a human retests. Only then is it done.

  1. 01Create a runYour agent calls create_test_run over MCP — or you use the dashboard — with the journey, the build and what a pass looks like. Nothing is tested until you fund it.
  2. 02Matching picks relevant humansPriced, then matched by device, language, domain and availability. Hard requirements gate who is eligible; relevance ranks the rest.
  3. 03They test and file findingsA Product Hand accepts via Telegram, runs the journey on a real device, and writes it up. Evidence is attached, not optional.
  4. 04A human verifier confirmsEvery bug is reviewed by an authorised human before it counts. The rest are rejected, marked duplicate, or filed as signal.
  5. 05Your agent pulls the queue and fixes itVerified bugs arrive as work your agent can act on — steps, evidence and the device it happened on.
  6. 06A human retestsThe same journey, run again by a person. Nothing closes itself, and a fix that did not land is reopened.

The one thing

AN AGENT CANNOT BE YOUR USER.

In their languageNot a translation of your copy.
On their deviceA real phone, not an emulator.
In their contextThe life they already have.

What comes back.

The bugSteps, video, and the exact device it happened on.
The personA handle, their domains, the languages they work in.
The verdictConfirmed by a second human before it counts.
The receiptWhat it cost, to the cent, in your own ledger.

What it costs

Priced per run, not per seat.

No packages, no subscription — credits are prepaid and spent as runs complete.

Effort

How long the journey actually takes to walk.

Scarcity

How rare that device, language or domain is in the network.

Severity

What a verified finding at that severity is worth to you.

Asked every time.