Privacy notice
Updated November 14, 2026. Applies to the Nabsun desktop browser and identifies the separate website's data handling. Nabsun is maintained by Naveen Alavilli. Questions: code@naveenalavilli.dev. Do not send passwords or private files in a public issue.
On your device
Nabsun stores browser history, cookies, site storage, bookmarks, settings, chats, attachments and activity records in your local application profile. Approved file edits can create local backups. Some credentials use operating-system protection; the profile and chat files are not a separately encrypted vault. Private browsing does not make online requests anonymous to websites or services.
The optional local model performs inference on your computer once installed. The desktop browser has no product analytics or automatic crash-report upload. This does not mean the whole browser is offline or that every configured integration runs locally.
Websites and routine connections
Websites you open communicate with their operators, receive network information such as your IP address, and may store cookies. Search queries go to the selected search engine. Website permissions are managed separately from assistant access.
Nabsun checks GitHub for a newer release at startup and periodically (currently every 24 hours). This request exposes ordinary connection information to GitHub; the update checker does not include chat text or local file content. Update downloads open in your default browser and installation is manual. Development setup downloads the engine/model and dependencies from their respective sources. Choosing the in-app Local AI download contacts Hugging Face and its approved content-delivery hosts. Nabsun verifies the pinned file before making it available.
Optional online assistants
Choosing Claude, OpenAI/Codex or another remote endpoint sends requests to that service. These may include your message, conversation context, tab information, attached page text, images, files, saved notes if enabled, and tool results needed for the task. A locally installed agent CLI can still use a cloud model. File access approval controls which files Nabsun can access; it does not guarantee that their contents stay on-device when a remote assistant uses them.
The provider's terms govern retention, processing and possible training use. Nabsun does not promise that remote requests have zero retention, zero training, or complete automatic redaction. To keep model inference local, use the downloaded local model or an independently verified local endpoint and avoid remote tools. Changing providers or disconnecting a folder does not recall data already sent.
Integrations and sharing
Enable only plugins, extensions and MCP servers you trust. They may have their own filesystem/network access and privacy practices. A local bridge connects authorized clients to Nabsun tools; the connected client may use remote services. The approval mode you choose affects when actions require confirmation. Review content before posting, uploading, submitting forms or sharing diagnostic logs.
Provider references: Anthropic, OpenAI, GitHub. Custom endpoints, websites and integrations have their own policies.
Control and removal
You can choose a local assistant, disable integrations, disconnect approved folders, and remove chats/site data through the available app controls. Local backups, downloads and files outside the profile may remain. Uninstalling Nabsun preserves its profile by default so an upgrade does not erase your information. Removing local data does not delete copies already held by external services; use the relevant service's deletion process for those copies.
The nabsun.com website
The marketing website uses Google Analytics on production domains. The test hosting site disables that integration. Website analytics is separate from the desktop application. Hosting and analytics providers receive ordinary web request information; Google's privacy policy applies to its processing. Links to GitHub, LinkedIn and other services take you to their respective policies. Requests you send to the maintainer are used to address the request; do not include more personal information than necessary.