seaLeye

A self-service front door to your organization's own certified cloud platform.

Version: 2.50.0, October 2026
License: AGPL-3.0, free to use, change and run yourself
Runs on: Node.js 20, AWS (ca-central-1 by default): CloudFormation for the starter modules, OpenTofu for an organization's own Terraform modules
Status: Working pilot, Phases 0 to 4 of the roadmap built. The author is stepping back; a maintainer is welcome.

What it is

Large organizations have built good platforms: libraries of infrastructure their security and audit teams have already approved. Developers still wait weeks for an environment, because someone from the platform team has to build each one by hand.

seaLeye is the front door to that platform. A developer runs one command in their app's folder, or opens the web portal. seaLeye reads the app, asks at most four questions in plain language, and writes a plan a person can read. On yes, it builds the environment from parts the platform team already trusts and hands back an address.

seaLeye learns that platform by reading the platform team's own library of Terraform modules: what each one is for, the rules for using it, and what it must refuse. It asks a platform engineer about anything it can't settle alone, and the catalog it writes has to pass a self-test before any developer plans from it.

If the app needs something the platform can't build, seaLeye says so, shows the code that shows the need, and builds nothing. Each refusal tells the platform team what to build next.

Try it in five minutes

The tests need no AWS account and no API key.

git clone https://github.com/GregKliewer/sealeye
cd sealeye
npm install
npm test

Running it for real takes an AWS account you're signed in to and an Anthropic API key. The README walks through it. To watch it learn a platform, point it at sealeye-sample-platform, a made-up organization's library, with an answer key, that every change to discovery is tested against. A hosted pilot is also running for now at seaLeye.online; open an issue to ask for a token.

What works, and what doesn't yet

Works

The interview, the plan and the refusal
Learning an organization's own platform from its Terraform library, with a platform engineer's answers
A self-test every catalog passes before it goes live, and every catalog kept with the run that wrote it and the answers it was given
Building from the organization's own modules, connected to each other, a database in the private network included
A gap report: what developers asked for that the platform can't build yet, and how often
A catalog that says which apps use each module, and at which release
Deploy, deploy again, pause, tear down
Settings in the organization's own store
Logs in plain words, the app's beside its front door's and its database's; performance, costs and evidence
Every model call through one gateway, with a spending cap on each run
A secret scanner, so a key written into the code never reaches the model
A command and a web portal that do the same things
A starter of four modules on AWS, for an organization with no library yet

Not yet

An organization's real code: tracing hosted in Canada, and a separate account per customer for learning its platform, come first
No path to production yet (Phase 5)
No sign-in: one token per person
One API process on one small server
A complete platform library of infrastructure-as-code and pipeline automation

A note from me

I built seaLeye in 2026 inside OpMo Advisory, my consulting practice. I'm closing the practice and turning my full attention to a new full-time role, so seaLeye won't be getting much of my time for a while. Rather than let it sit in a drawer, I'm putting it out here for anyone who wants to use it, fork it, learn from it, or contribute to it.

The code is yours under AGPL-3.0. The name stays with the original, so a fork needs a name of its own (NOTICE has the details). If you'd like to carry seaLeye itself forward as its maintainer, open an issue and tell me. I'm open to handing it over.