Configuration

Where config lives

Atlas looks for the first of these in the project root: atlas.config.json, .atlasrc, .atlasrc.json, or an "atlas" key in package.json. CLI flags always win over the config file.

{
  "include": ["src/**/*.{ts,tsx}"],
  "exclude": ["**/*.stories.tsx"],
  "outDir": ".atlas",
  "sharedLayers": ["shared", "common", "ui"]
}

{
  "name": "my-app",
  "atlas": {
    "exclude": ["src/legacy/**"]
  }
}

Keys

What is scanned by default

Without an include, Atlas scans **/*.{ts,tsx,js,jsx,mjs,cjs}. The built-in excludes cover node_modules, dist, build, .next, out, coverage, .atlas, declaration files, and *.test.* / *.spec.* / *.stories.*. Your own exclude entries are added on top of that list, so you only need to name what's specific to your project.

Set ATLAS_VERBOSE=1 in the environment for verbose logging.

Monorepos

Nothing to configure. npm, yarn and pnpm workspaces, Turborepo and Nx are auto-detected from the workspace manifests, and each workspace's path aliases resolve on their own — including split tsconfig setups where a base config is extended by per-package files. Routers are resolved per workspace too, so two apps in one repo don't blur into a single route table.

Shared layers & architecture

sharedLayers tells the architecture pass which top-level folders are the common layer every module may import from. Anything else that reaches across module boundaries is reported as a violation, with a recommendation, in architecture.json and the dashboard.

Cache

The incremental cache is on by default and lives in .atlas/.cache/. It keys on file contents and the Atlas version, so upgrading the tool invalidates it automatically. Disable it per run with --no-cache, force a clean scan with --reanalyze, or turn it off permanently with "cache": false.

Plugins

A plugin contributes extractors (new asset kinds or frameworks) and can post-process the final result. Register it in plugins; see Architecture → Plugins for the interface.