Our approach to privacy in an AI editor
Why privacy mode is the default and how zero data retention works for enterprise teams.
An AI editor sees everything: your code, your comments, your half-finished ideas. That access is exactly why the defaults matter. Here’s how we think about privacy at Arshas, and what we actually guarantee.
Private by default, not by upgrade
Privacy mode is on by default for every account, including free ones. With it enabled, code sent for completions, chat, and agent runs is processed in memory, used only to serve your request, and never stored or used for training. We think making privacy a paid feature gets the incentives exactly backwards.
The codebase index — the thing that makes Arshas smart about your repository — stores embeddings and symbol metadata only, never raw source. You can verify privacy mode is active at any time from the shield icon in the status bar.
Zero data retention for enterprises
For teams with contractual requirements, Enterprise plans add zero data retention: request payloads containing code are never written to disk anywhere in the pipeline, including at our model providers, and the guarantee is backed by your DPA rather than a settings toggle. Even operational logs for ZDR traffic exclude code content.
Admins can lock privacy settings workspace-wide, connect SSO, and pin index storage to a region. Security teams get audit logs for every authentication and admin event.
The principle
Our rule is simple: we should be able to explain, in one paragraph and without lawyers, what happens to your code on every request. If a feature can’t meet that bar, it doesn’t ship. Your code is your company’s crown jewels; treating it that way is the whole strategy.