You built the app. Get it into users’ hands.

In the console

Starting from a new organisation with a Node app and a Postgres database, pick a repository. Ankra reads the code and opens a pull request with what it takes to build and deploy it.

The checks run and the release stops: the image gate found a critical vulnerability in a package the app pinned, and nothing is published. Press Fix with Ankra and send it in Agent mode.

Its first idea changes one file on the wrong branch, which would break the lockfile, and it cannot do that without you. Decline and say why. It then runs the real package manager in a sandbox, regenerates the lockfile, and proposes two files on the right branch. Approve it, Ankra commits, and the same gate passes.

Merge the setup. The image that pull request built and scanned is published with no rebuild. Choose staging, make it public, and Ankra sets up Postgres on that cluster and deploys. The app answers on its own address with a real certificate.

Two switches give every pull request a preview and an AI review. Merging moves staging by itself. Production waits for a person, who chooses the version staging proved and confirms. Then Slack is connected once, and a pull request that breaks a test tells the team which run and which step failed.

With the CLI

The same path from one terminal, starting with two clusters created with one command each. ankra application ship registers the repository, opens the pull request, and stops because it is not merged yet.

The pipeline runs in your own cluster, the gate blocks the same vulnerability, and the block reaches Slack with the package and the fix. Ankra AI, in Agent mode, regenerates the lockfile in a sandbox and queues the commit. It has not run until you confirm it.

After the merge, the same command picks up where it stopped, sets up Postgres, and prints the address. Production waits for a person and gets the exact image staging ran, by its tag. Adding a book through the API and reading the page back shows a real database over HTTPS.

With Claude Code

A real Claude Code session, connected to Ankra over MCP with a scoped, expiring token. Ask for production, and tell it to stop if anything blocks. It registers the repository, follows the pipeline, and reports the critical vulnerability: nothing merged, nothing deployed.

Told to fix it, it regenerates the lockfile, commits to the right branch, and the gate passes. It stops at the merge, because that is still your call. Once approved, it deploys to staging with its database and checks the address over HTTPS.

On an everyday change its first wording broke a test, and it fixed its own change rather than the test. Production needs a person, so it says what it would deploy and where, and waits. When told yes, Ankra refuses because production does not have the database operator, and Claude does not widen what you approved. With the add-on in place, it deploys.