Skip to content

WorkZeasy RAG

In-house tool · AI · since 2025

Our coding agents first ask what already exists.

Zeasy RAG is our knowledge store for code and documentation from our projects. If an agent finds a finished building block there, it doesn’t build it again.

Graphic: a dark grid of many blank index cards. Six of them glow violet and magenta, and a line runs from each to a bright point on the right. That’s how an agent searches the store and gets the matching hits back.

Starting point

Without central knowledge, a coding agent starts from scratch every time.

Many of our projects need the same building blocks. One example is sending email: it’s part of almost every website we build.

Either you explain everything to it again, or you show it another project to copy from. Then it has to search, read, understand, evaluate and rebuild. That takes time and costs computing power. In the end, every version is a little different.

What we built

  1. Code and documentation in one place.

    Code and documentation from our projects sit in one place and can be searched there. When someone changes something, the store is up to date shortly after.

  2. Search comes before writing.

    Before an agent builds a building block from scratch, it asks whether one already exists. The answer comes with its source: which project, which file.

  3. New projects start with the latest version.

    When a building block gets better or more secure, we change it in one place. The next projects get the new version automatically.

  4. Vulnerabilities across all projects in view.

    The store knows the libraries of every project. If a known vulnerability shows up anywhere, we see which project is affected.

Result

That way, the websites we build get building blocks that already run on other sites.

For sending email, the agent takes the finished building block from the store and doesn’t write it again. That saves time and computing power.

Today we also build new AI tools on top of the store.

For your company

The same principle works with your knowledge: contracts, manuals, tickets, code. You decide what goes in and who may query it.

We plan the scope with you.

In a similar spot?

We build this kind of system under AI, automation & modern development. AI ticket routing shows how we train a behavior into a model instead.