Their own computer. Never yours.
Every project's agents get one always-on computer of their own. They build, test and ship code there, browse the internet in a real browser, run your app behind a preview link, and give every job its own isolated space. Staff laptops stay out of reach.
One ticket, one isolated workspace, watched line by line.
A design arrives. The agents pick up the ticket, start a fresh workspace on its own branch, build the app and run your checks. By the next stand-up there is a working preview link on the ticket, with the results beside it.
- Every job starts clean, so two tickets never trip over each other.
- The preview link and the check results land on the ticket.
- Every job in flight is on one screen, with retry and cancel.
- Try a new agent here first. A run that goes wrong is thrown away, and no one's machine is touched.
Build
Your app is built the way your repository says to build it.
Format and lint
Style slips and common mistakes are caught before a person looks.
Test
Your suite runs, and the result goes on the ticket.
Fix
When a check fails, the quick automatic fix runs first. If that is not enough, an agent repairs it and runs the check again.
See it working before you read the code.
When the change is something people will look at, a passing suite is only half the answer. So the agents run the app on their computer and show you what they built.
A preview link for every change
The app runs in the job's own workspace, and you open it from the ticket. Only people in your workspace can reach it.
Screenshots as proof of work
For a ticket that asks for visual proof, the agent opens the running app in a real browser and attaches what it saw, so review starts from the screen your customer will get.
Behaviour scenarios, green
Given, when, then. Turn on BDD for a ticket and the agents write its scenarios first. The job runs them with the rest of your tests, so intent and code stay in step.
They browse the internet like a person. You can watch, and take over.
Every Cauldron has a real browser. Agents open websites, read what is on the screen, click, type and take screenshots, so a supplier portal or an old admin tool with no API is still within reach.
Browse like a person
Open a page, read it, fill in a form, download a report. The agent works the site the way your team would, on its own computer.
It asks before it acts
Submitting a form, paying, or opening a site it has not visited before waits for a one-tap approval from the person the agent is working for. What it types is never stored or shown.
Watch live, take over
See the agent's screen from chat, the assistant or the Cauldron page. Take over when you want to steer, then hand it back.
No separate pipeline service to wire up and babysit.
Describe your pipeline in one short file in the repository: the stages, what each one waits for, the secrets it needs and where it deploys. It runs on a push, a pull request or a tag, on the same computer your agents already work on.
- Agent changes and people's changes get the same checks.
- Stages run side by side where they can, and a failure stops whatever depends on it.
- A secret reaches only the stage that asks for it.
Deployments carries it the rest of the way.
When a person approves and merges the change, Deployments ships it to the environment you pick, and keeps the history so you can go back.
Deployment history
Every release, tracked.
One-step rollback
Go back to the previous release in one step, blue/green switches included.
Vaults
Secrets and settings for each environment, kept out of the code.
Provisioning wizard
Set up a managed platform or a server, one guided step at a time.
Service topology
A live diagram of how your services connect.
Snapshot diffs
See exactly which settings changed between releases.
Give your agents a computer of their own.
Connect a repository, and the next ticket runs in its own workspace with a preview link waiting when it is done. That's the juju.
