Skip to main content
Gliss has three parts, and your app’s code passes through none of them on its way to your cloud.

​
From idea to live app

1

You ask your AI for an app

You describe what you need. Your AI writes the code and a short gliss.yaml that asks for what the app needs, such as a database or sign-in, from the catalog your IT team publishes.
2

Gliss creates the repository

You sign in with your company account. That is the only account you need: no GitHub seat, no AWS access, no Docker. Gliss creates the repository, and your AI works on your own branch.
3

Your company's CI builds it

Each push builds the app in your company’s CI. CI holds no stored secrets: it proves who it is with short-lived tokens. Unchanged parts are not rebuilt or retested, which keeps every round fast.
4

Gliss checks the request and signs a release

Gliss checks gliss.yaml against the catalog. Anything off the menu is refused with a message your AI can act on. Otherwise Gliss signs a release: a statement of exactly what may run, where.
5

Your cluster applies it

The Gliss agent in your cluster receives the signed release over a connection it opened itself, verifies the signature and applies it. Gliss cannot log in to your cluster. It can only sign releases that your cluster checks.
6

Preview, Beta, Live

You try the app in a private Preview, share it as a Beta, and press Publish when it is ready. Live runs exactly the build that passed Beta.

​
When something goes wrong

Your AI does not need you to copy error messages. After each push it receives a diagnosis: what failed, whether it is the app’s problem or the platform’s, and what to do next. If it is the app’s problem, your AI fixes it and pushes again.

​
What Gliss can and cannot do in your account

Gliss can sign releases, which your cluster applies only if they verify and your IT team’s policy allows them. It cannot log in to your AWS account or cluster, read your secret values, or run arbitrary commands. If Gliss cloud is unreachable, your apps keep running. See the security model for details.