How Atlarix reads your codebase

Atlarix does not git-scan or index your repository when you open it. There is no embedding step and no background crawl, so opening a large monorepo is as fast as opening a small project — and nothing about your code is precomputed and stored.

The cost of that choice is that the agent explores on demand, which is what the two primitives below are for.

Answers "where is this file?" using patterns like **/*.test.ts or src/**/auth*.

Results are sorted by modification time, so the files you have been working in come first. Use it when you know roughly what something is called but not where it lives.

Answers "where is this used?" — a ripgrep content search with full regex support.

Two things make it more useful than a plain search:

• Results are relevance-ranked, so the most relevant files come first rather than in directory order.
• Each hit is annotated with its enclosing function or class, so the agent can tell a definition from a passing mention without opening every file.

Once a file is located, the agent reads specific ranges rather than whole files where it can — large files are paged, never silently truncated.

This is why a good question narrows the search: the more precisely you point at an area, the less the agent has to read to answer well.

Type @ in chat to focus the assistant on part of your open project.

Mention a file and its contents come along. Mention a folder and the assistant gets the list of files inside, then reads only what it needs within that area.

It is a precise alternative to pasting code, and it avoids pulling in your whole project.

For a broad question spanning several areas, the agent can fan out read-only scouts in parallel — one per area — and synthesise their reports. See the sub-agents guide.