Azilos

Azilos upgrades a Node.js or TypeScript repository and proves the result still works. It runs on your own servers. You give it a Git repository, and it gives you back a branch, a report and, if you want one, a pull request.

What a run does

1. Reads the repository. Azilos finds the package manager (npm, pnpm or Yarn, including workspaces), the dependencies and the Node.js version the repository declares.
2. Plans the upgrade. It lists each outdated or vulnerable dependency with its current and target version, marks major versions as breaking, and plans the move to the Node.js version your install sets as its target. You can exclude any upgrade, or set a policy once: patch and minor versions only, or major versions too.
3. Edits a branch, never your default branch. The version changes and a regenerated lockfile are committed to a new branch.
4. Checks the branch against the original. Azilos installs, builds and tests both the original commit and the new branch in containers, with the repository's own scripts, and compares them step by step.
5. Proposes a fix when something breaks. If a step that passed before now fails, Azilos asks an AI model for the code change that fixes it. Nothing is committed until a person on your team approves the proposal. An approved fix is committed to the branch and checked again.
6. Reports what happened. The report lists every upgrade, every approved edit with its reason, and what passed and failed on each side. With pull requests turned on for the project, Azilos opens one on GitHub with the report as its description.

The glossary explains the report's terms.

Where your code stays

Azilos is self-hosted. The API, the dashboard, the database and the containers that build and test your code all run on a host you control. Your team signs in through your own identity provider.

One thing can leave your network: when a check fails and your team has consented to AI fixes, the failing files and the end of the build output are sent to Azilos Cloud, which forwards them to the AI vendor. Well-known secret shapes are removed first, and a file that holds one is not sent at all. What leaves your network describes exactly what is sent, and when.

How AI is paid for

The AI that proposes fixes is sold as prepaid credits on Azilos Cloud. You buy a credit pack, create an API key on your account page and give it to your install. Each AI request is charged at the token counts the vendor reports. You can set a daily spending limit on each key and ask for an email when the balance runs low. When the credits run out, no new run starts, and everything your install has already produced stays readable.

Pricing explains credits and links to the current prices. Manage credits and keys on your account page.

Install it

Install with Docker Compose brings up the whole stack with one command. The installation guide covers a team's own host: settings and secrets, backups, moving to a newer version, your identity provider, and the first run.

Azilos is in pre-release. The guides say which steps have been run as written and which have not.