Privacy Policy

Effective date: October 1, 2026 · Applies to the browsa browser extension and this website

No browsa account, no telemetry, no developer relay server. The extension stores settings and conversations locally. When you use AI features, your browser sends the relevant content directly to the model or Agent service you configure.

What data is processed

browsa processes the following data to provide reading and chat features. It does not send usage records, page content, or personal information to the developer, has no analytics, and does not sell user data. The source is publicly auditable under the MIT license at github.com/xiaohuzai/browsa.

Pages and files: the current page's URL, title, readable text and subtitles, plus documents, screenshots or images you attach. Content scripts on supported sites observe loaded data when the page loads so it can be extracted later; attachments provide conversation context.
Conversations and settings: your questions, model replies, attachments, saved sessions, model and Agent configurations, and interface preferences.
Authentication information: API keys you enter authenticate with their respective services; on-demand platform media retrieval may use that platform's login context.
Audio and video (optional): media you choose to transcribe or analyze, used to produce transcripts, notes or audiovisual analysis.

Where data lives

Chat history (including saved attachment images), saved sessions, service configurations, API keys and preferences stay in chrome.storage.local on your device. Temporary runtime state uses memory or chrome.storage.session. browsa provides no cloud sync.
You can delete messages or sessions and change configurations within the extension. Uninstalling removes its browser storage; files you exported must be deleted separately.
Data sent to a model, Agent or analysis service is processed and retained according to that service's settings and privacy policy. Deleting a local session does not delete copies held by the service.

Where network requests go

Models and Agents: conversations, questions and relevant attachments go directly to the address configured in Settings, without passing through a browsa developer server. API keys are transmitted to authenticate with their service. Connection checks also contact the configured service. An initiated turn may continue after you switch sessions.
Local parsing and model requests: PDFs and Office documents are parsed into text and images on-device; those results can be attached and sent to your configured model or Agent. Recognized credentials in page-context URLs, such as tokens, passwords and signatures, are masked before sending. This is not a comprehensive filter for sensitive page text and does not rewrite URLs used to fetch resources.
Site extraction: besides reading already-loaded data, browsa makes on-demand requests to the current site and its APIs for text, comments, subtitles, documents or media, including raw code files and arXiv paper metadata. Requests may use the browser's existing login context. arXiv metadata queries send a paper identifier, not conversations.
Audiovisual analysis (optional): your browser retrieves media from its platform or CDN on demand, with that platform's Cookie or Referer headers where required. After local transcoding, it sends audio or analysis inputs to your configured transcription or analysis service. This processing occurs only when you enable and use the relevant features.
External reply resources: displaying reply images contacts their source. Protein structures requested by identifier contact structure databases such as RCSB or AlphaFold and send the corresponding identifier. These requests do not send the full conversation.
Website: web fonts come from Google Fonts, which your browser contacts directly for font resources. The website adds no analytics or tracking scripts. External links lead to sites with their own privacy policies.

Permission usage

<all_urls> / tabs / scripting: read and attach pages, screenshots or documents, inject extraction scripts as needed, connect to configured services and seek videos when you click a timestamp. Supported-site observation scripts and selection tools load with the page. They do not upload your entire browsing history to the developer.
cookies: read the platform's login context when audiovisual analysis retrieves its media, to construct required headers; not sent to the developer.
declarativeNetRequest: temporary session rules add Referer-type headers during media retrieval; removed after use and cleared when the browser session ends.
webNavigation: detect navigation while the side panel is open and update current-page information.
storage / unlimitedStorage: save the local settings, conversations and attachments described above.
sidePanel / contextMenus: display the chat side panel and context-menu actions.
alarms: runtime maintenance such as clearing temporary stream state; not telemetry.

Third-party services

The browsa developer operates no backend for receiving extension data. Your chosen model, Agent and analysis services, and content-source sites, each have their own privacy terms. With an on-device model, model requests can stay on your device. Whether a local Agent calls external models or tools depends on its own configuration.

Children

browsa does not request your age or target children for data collection. The age requirements and privacy policies of configured third-party services still apply.

Changes & contact

Material changes will be posted on this page with an updated effective date. Questions? Open an issue at github.com/xiaohuzai/browsa/issues.