# Knowledge you can build on.

Your knowledge should grow with your work. So should the software you use to understand it.

## Give knowledge a structure you can check.

A codebase gives an agent more than text: types, relationships, conventions, and feedback when something breaks. A knowledge base should offer the same.

Typed files make expectations explicit. A claim can require a source. A reference can point to a particular kind of object. Structural checks catch mistakes before another session builds on them. Whether the source supports the claim still takes judgment.

## Build the tools as you build the knowledge.

Your questions change as your understanding grows. You should be able to change the environment with them.

That is what malleable means to us: reshaping the knowledge model, the agent’s methods, and the interface together. When a view or capability is missing, you and your agent can build it. Our app uses the same primitives and SDKs available to you.

## Give each session something lasting to build on.

Conversations end. The work should carry forward.

Repos bring knowledge together with the types, skills, and tools used to work on it. They can depend on one another, letting you reuse a method or vocabulary across projects. Different agent sessions can inspect and revise those durable files, continuing from what came before.

## The art of shadows.

A graph can grow beyond what you can hold in your mind at once. Each projection reveals a part: a document, a map, a set of relationships. Moving between them helps you understand the whole.

Ars Umbris takes its name in homage to ars memoriae, ars combinatoria, and the tradition of building structures to think with. The art of shadows is how we think about shaping the views through which humans and agents understand their work together.

## Your thinking should outlast the tool.

Your repos remain files on your machine. Keep them in Git, open them elsewhere, and keep reading them if you stop using Ars Umbris.

## Start with something you want to understand.

Build the knowledge together, then build the tools you need around it.