Privacy

Ask runs private document Q&A in your browser with filename/chunk citations. Inference, retrieval, and enabled memory stay on this device after model assets are available. Ask does not collect product analytics or telemetry.

What stays local

When network features are off, Browser-provider prompts and responses, conversation history, document chunks, and RAG vectors stay under the app's browser origin. The original file is not uploaded for indexing. Ask computes a SHA-256 digest of its bytes locally for cache identity and duplicate detection; the digest is not sent to a server. Removing or clearing an attachment waits for its IndexedDB cache deletion to commit before the UI reports success.

For an inspectable cited answer, Ask stores the exact retrieved passages used for that answer with its conversation in Chat History. Removing an attachment or clearing the RAG Vector Cache does not rewrite old answers; clear Chat History to erase those evidence snapshots.

Ask saves and retrieves semantic memories only while Memory is enabled. Turning Memory off leaves existing entries stored until you delete them. Model weights use browser-managed caches; browser eviction, clearing site data, or clearing Model Weights removes them. Cloud API keys use sessionStorage rather than localStorage or IndexedDB, although crash and session-restore behavior varies by browser.