What arcy init and arcy analyze read and send
A breakdown of what the CLI scans and transmits during setup.
arcy init and arcy analyze read your source files locally, with an AST parser, to find your routes and generate your .arcy/ knowledge graph. They do not identify or write anything about interactive elements. Raw file content never leaves your machine, under any command. Always skipped: node_modules, build output, .env*, *.pem, *.key, symlinks, and oversized files.
| Step | Sends | Where |
|---|---|---|
arcy init install step | Package name only | npm registry |
arcy analyze | Nothing | N/A |
arcy push | Your .arcy/*.yaml content (routes, labels, any hand-authored sections) | ARCY backend |
| Enrichment (automatic) | The structural map only: routes and labels | ARCY's model provider |
The only network call that touches a model, analyze/enrich, receives just the structural map already in your .arcy/ files, never your JSX or logic.
Element-level detail (what's on a page, and the elements Copilot/Autopilot can act on) is not part of the CLI at all. It comes from the SDK's passive autocapture, which is a separate data path covered in What ARCY collects.
Enrichment tenant gating: Auto (default for most plans) runs enrichment immediately. Consent-required (default for enterprise-tier workspaces) skips the model call until you grant consent, returning deterministic YAML stubs instead. See Zero Data Retention for how AI request content is handled after a call completes.