Available NowCurrent extension architecture

Local first.Cloud by choice.

Typing triggers local analysis inside the extension. Prompt text reaches BeforePrompt’s backend and configured AI provider only after the user chooses a cloud-backed action.

One-page data flow

Four boundaries, with the person still in control.

Active editor

The extension reads the unsent draft on a supported AI page.

Local check

Clarity and sensitive-data checks run in the browser without a cloud request.

Deliberate action

The user chooses Improve, Clarify, Structure, Refine, or another labelled cloud action.

Temporary processing

The backend authenticates the request and sends the required content to the configured AI provider.

Storage boundary

Raw prompt history is off by default.

Operational events can store action type, site, score, length, latency, token usage or estimate, status, and timestamp without raw prompt text. Original and suggested prompt text is stored only when the user turns on prompt history.

Team analyticsAggregate usage and coaching signals do not expose raw prompts to administrators by default.

Optional historyUsers can delete individual history items or clear all saved prompt history.

Send controlBeforePrompt can write an accepted suggestion into the editor; it never presses the AI site’s Send button.

Permissions

Narrow host access.

Requested

Chrome storage, the four supported AI hosts, and beforeprompt.com for authentication and user-requested APIs.

Not requested

No all-sites permission, no background capture of unrelated browsing, and no automatic cloud rewrite while typing.

Complete terms