yeahhe365/Gemini-Nexus

Gemini Nexus – AI‑powered Chrome extension

What it is – Gemini Nexus is a Chrome MV3 extension that embeds a conversational AI assistant directly into the browser. It can talk to Google Gemini (both the web UI and the official API) as well as a wide range of OpenAI‑compatible services (OpenAI, Anthropic, DeepSeek, OpenRouter, Qwen/DashScope, Zhipu, etc.). The extension adds a side‑panel, a floating toolbar, and a set of browser‑control tools so the model can read pages, click buttons, fill forms, take screenshots, and even run custom JavaScript.

Why it matters – Instead of opening a separate chat window, the assistant lives inside the browser and can act on the current tab. This makes it possible to use LLMs for on‑the‑fly translation, summarisation, code generation, or fully‑automated browsing tasks without leaving the page.

---

Core capabilities

Feature	Description
Multi‑provider support	Switch among Gemini Web (no API key, reverse‑engineered), Gemini official API, OpenAI‑compatible APIs, Anthropic, DeepSeek, OpenRouter, Qwen/DashScope, Zhipu, etc. – each with its own base URL, API key and model IDs.
Native browser control	Through Chrome DevTools Protocol the model can issue commands like navigate_page, click, fill_form, take_snapshot, evaluate_script, etc. Controlled tabs are grouped for visual feedback.
Floating selection toolbar	When you highlight text, a toolbar appears offering translate, summarise, explain, grammar‑fix, or insert the result back into a form.
Image & screenshot input	OCR/translation of selected image regions, full‑screen or window capture, and a floating “analyse image” button for page images.
Side‑panel UI	Built with the Chrome sidePanel API; fast access, full‑text history search, markdown/LaTeX/code rendering in a sandboxed iframe.
Context management	Summarisation compression and turn‑trimming to keep token usage within model limits.
External MCP tool integration	Connects to local or remote Multi‑Tool‑Calling (MCP) servers via SSE, HTTP streaming, or WebSocket, exposing custom tools to the LLM.
Watermark removal	Strips embedded watermarks from Gemini‑generated images (use at your own risk).
Temporary Gemini Web chats	Allows using Gemini Web without polluting the user’s official chat history.

---

How to get it

1. From a released build – Download the ZIP from the Releases page, unzip, then open chrome://extensions/, enable Developer mode, click Load unpacked and select the extracted folder.
2. Build from source –

   npm install
   npm run package:extension   # creates a ready‑to‑load folder in artifacts/chrome-extension
   

   Load that folder the same way as above. (npm run build only produces the UI bundle, not the full extension.)
3. Publish – The repo includes a script (npm run publish:chrome-webstore) that uploads the packaged ZIP to the Chrome Web Store using credentials stored in a local .env.chrome-webstore file.

---

Architecture snapshot

• Frontend – Vite + TypeScript UI compiled into dist/ and injected into the side‑panel and floating toolbar.
• Background – Handles provider routing, token storage, and the Chrome DevTools‑based control loop (background/control/).
• Content scripts – Injected into web pages to capture selections, add the floating image‑analysis button, and perform DOM actions via the Accessibility Tree.
• MCP bridge – Optional proxy (e.g., MCP‑SuperAssistant) translates extension calls to local stdio‑based tool servers.
• Provider drivers – Located under services/providers; each driver knows how to format requests for its target API (e.g., web.js for Gemini Web, official.js for Gemini API, anthropic.js for Claude, etc.).

---

Risks & caveats

• Gemini Web reverse‑engineering – The extension extracts internal RPC tokens from gemini.google.com and mimics the web client. This likely violates Google’s Terms of Service and could stop working if Google changes the internal API.
• Watermark removal – Stripping attribution from generated images may breach Google’s usage policies.
• API keys – All keys are stored only in the extension’s local storage; they are never sent to third parties beyond the chosen provider.
• Responsibility – The authors provide a disclaimer that users assume any legal or policy consequences of using the reverse‑engineered features.

---

Who might use it

• Power users who want a conversational assistant that can act on the current web page without switching apps.
• Developers looking to experiment with LLM‑driven browser automation or tool‑calling workflows.
• Researchers interested in testing multi‑provider LLM integration and the limits of reverse‑engineered access to Gemini Web.

---

License – MIT (open source).

---

Gemini Nexus brings a native AI layer to Chrome, unifying many large‑language‑model providers and giving them direct control over the browser for tasks ranging from simple text assistance to full‑blown autonomous browsing.