How it works

How a test runs

Once your agent has its images, every test follows the same steps on the same kinds of machine.

Step by step

The same six steps each time. The prompt tells your agent to give each run its own project name, so runs side by side keep to their own machines.

  1. Build an image with your software installed, from an Endiafile
  2. Start the machines endia compose up --wait
  3. Install or start whatever the image leaves to start
  4. Act and check over the network, inside the machine and on its desktop
  5. Collect evidence screenshots, files, logs and event-log entries
  6. Remove the machines after every test, passed or failed

Features

Desktop control

Your AI controls each machine's desktop: it reads windows, clicks, types and takes screenshots. With one registry line in the image's recipe, it answers elevation prompts too.

Fleets

Your AI can test a server and its clients talking to each other, on a private network described in one Compose file.

Images

Your AI writes image recipes in Dockerfile syntax, layering your runtimes and software over a Windows image you prepare once.

Agent access

You authorise a client by generating a client certificate, using either the desktop app or endia context export at the CLI. You can revoke the certificate at any time.

Connecting from a Docker sandbox ›

On the PC
# you, on the PC, when the agent asks
sbx policy allow network --sandbox <sandbox> localhost:7443
endia context export <sandbox>

# you, on the PC, to end its access
endia cert rm <sandbox>

Ready to hand it over?