01Scope

"We" refers to the maintainers of the NeonAgent open-source project. "The software" is that Chrome extension; "this site" is the static page that introduces it.

This is a personal open-source project — offering an extension and optional account and model proxy services. This document describes what the software and this site actually do with data. It is not legal advice; for anything consequential, assess it yourself or consult a professional.

02Open-source software: BYOK and built-in models

The extension supports bring-your-own-key providers and an optional built-in model. Custom providers receive requests directly; built-in requests pass through your configured NeonAgent backend to the official DeepSeek API.

The backend stores email, membership, hashed session credentials, request IDs, timestamps, models, statuses and token counts for authentication, accounting and administrator reconciliation. It does not store prompt or response bodies in the database. QQ mail delivers login codes. The software is open source under the MIT license.

03This website

This site is fully static: no cookies, no analytics, no advertising, no third-party tracking scripts. It uses system fonts and requests no external font service.

• The interactive demo is preset. The "agent reads the page for you" animation is hard-coded sample data. It makes no model request, and nothing you type or click is uploaded.
• The product page external request is a call to GitHub's public API for this project's star count, used for the number on the button. That request exposes your IP address to GitHub and is covered by the GitHub Privacy Statement. If it fails, the page falls back to a built-in number.
• The product page stores preferences locally — interface language and light/dark theme — kept in your own browser's localStorage and never sent anywhere. Clearing browser data removes them. The account console also stores login credentials and calls the backend for email verification, login and usage queries. Clearing local data does not delete server records.

04What the extension processes, and why

To provide chat, image analysis, page translation and browser tasks, the extension processes text you enter, images and reference files you upload, and the page text, URLs, screenshots, tool results and conversation history involved in a task. When site-specific features are enabled it may also read question content and API responses from that site. What is sent depends on which feature you use.

Please note: page content is structurally simplified before it is sent, but that does not guarantee removal of sensitive information — such as visible text or values already filled into a form. Be aware of what is on the page before you start a task.

05Where data is sent

Model requests send the relevant content to the API endpoint you configured, authenticated with the API key you provided. Built-in model requests instead pass through your configured NeonAgent backend to DeepSeek using a server-held API key.

How each provider stores and uses it is governed by that provider's own terms and privacy policy — review them before configuring. If you use an official DeepSeek, Kimi, MiniMax or similar endpoint, their usage policies apply as well.

Also note that once switches such as automatic translation, scheduled tasks or auto-solving the current page are enabled, the extension may process content without a separate click for each request. These are off by default and can be turned off at any time.

06Permissions the extension requests

Why a broad host permission such as <all_urls> is required

This is a general-purpose tool, not one built for a single website. Which pages it may act on is your choice, and no fixed domain list can be written in advance — so broad host access has to be declared. There are only three uses:

• Reading pages. When you ask it to summarize, answer a question, solve a quiz or analyse a page, it must read that page's text and structure to have anything to give the model.
• Translation. Full-page and selection translation act on "whatever page you are looking at right now"; without access to arbitrary pages, cross-site translation is not possible.
• Page interaction. For the agent to click, fill, scroll and submit on your behalf, it must perform those actions on arbitrary pages.

One further reason: the model endpoint is one you enter yourself (bring-your-own-key), so its domain cannot be enumerated ahead of time — and reaching that address also requires broad host access.

What the permission does not do: opening a page does not upload it. Page content is sent to your configured endpoint only when you start a conversation, a translation or a task; the switches that could run without a per-request click (automatic translation, scheduled tasks, auto-solving) are off by default. Built-in model requests pass through the configured backend.

Page-behaviour script: always injected, dormant by default

So that switches such as "unlock right-click" take effect the moment you turn them on — which requires running before the page's own scripts — the extension injects a small script into every page at the earliest point of loading. While those switches are off (the default), it only registers listeners that are switched off; it takes over the relevant events only after you enable a switch yourself. It serves three switches: unlocking right-click and copy restrictions (restoring the browser's default context menu and copy/paste), preventing a page from forcing fullscreen, and blocking common developer-tools detection signals.

07What is stored locally

API configuration, keys, conversation records, memories, skills and scheduled tasks live in the browser's extension local storage. Note that:

• This is not an encrypted credential vault. API keys are kept as extension local storage, and anyone with access to that browser profile may be able to read them.
• Some conversation flows may save image previews alongside records. Do not assume every image is deleted when the panel closes.
• Local history is not synced wholesale. Built-in model calls send the content needed for the request and create account and usage records. The extension includes no advertising trackers.

08Script skills and local commands

Script skills let you or the model add JavaScript that the extension executes, and scripts can reach the network through the extension's proxy. What a script does depends entirely on its content — use only scripts you trust, and check what they intend to do first.

When local commands are enabled, the extension connects to the WebSocket address you configured and exchanges commands and execution results with that service. Connect only to services you control or trust.

09Your controls and deletion

• Turn off automatic processing: translation, scheduled tasks, auto-solving.
• Delete conversations, memories, skills and scheduled tasks from the side panel, and remove saved API configuration.
• Uninstalling the extension clears all of its local storage in that browser.
• Data already sent to an external model provider must be handled with that provider — we cannot delete it for you.

10Usage limits and your responsibility

• AI output can be wrong. Answers, translations and action decisions may be inaccurate. Verify before relying on them in exams, financial, medical or legal contexts.
• Automated actions are your responsibility. When the extension clicks, fills and submits on a website, you must comply with that site's terms of service, its robots rules and applicable law. Do not use it to bypass access controls, inflate traffic, or for other abusive purposes.
• Be careful with sensitive data. Unless you have accepted the data policy of the provider you configured, do not use it on personal, confidential, financial or medical information.
• Third parties are independent. This project is not affiliated with or endorsed by any model provider; your account, quota and compliance obligations are between you and that provider.

11Disclaimer

The software and this site are provided "as is", without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose and non-infringement. To the maximum extent permitted by applicable law, the maintainers are not liable for any direct, indirect, incidental or consequential loss arising from the use of, or inability to use, the software — including data loss, service interruption or legal consequences.

The project is open source under the MIT license, Copyright (c) 2026 NeonAgent.

12Age requirement

The software and this site are not intended for anyone under 13, or under the minimum age of digital consent in your jurisdiction.

13Changes and contact

This document may be updated as the software changes; the date at the top will change with it. For how features behave, the privacy notice inside the extension is authoritative.

Questions or data-related requests: please open an issue on the project.