Codebase indexing
The semantic index that lets Arshas reason about your whole repository.
Everything smart in Arshas — chat answers, agent plans, next-edit predictions — is grounded in the codebase index: a semantic map of your repository built the first time you open a project and kept fresh incrementally as you work.
What gets indexed
The index covers symbols, definitions, references, and the embedding of every file, so the model can retrieve by meaning rather than by string match. Asking about “rate limiting” finds the throttle middleware even if the word “rate” never appears in it.
Files matched by .gitignore are skipped by default, and you can exclude additional paths with an .arshasignore file at the repository root.
Performance and privacy
Indexing a mid-size repository takes seconds; large monorepos index in the background while you work. Only embeddings and symbol metadata are stored — with privacy mode enabled, your source never persists on our servers, and enterprise plans can pin indexing to a dedicated region.