Shizue 发布开源 Chrome 扩展，把 LLM 接入浏览器日常浏览

👋 Intro

Shizue is a Chrome extension designed for integrating Large Language Models (LLMs) into the daily Browse workflow. The project aims to enhance browser interactions with LLM-driven features, conceptually similar to how tools like Cursor augment code editors.

Shizue provides a free, open-source alternative to commercial services (e.g., Sider) that restricts access to newer models or handy services behind additional paywalls. Sign in with ChatGPT to use your ChatGPT plan, or bring your own API keys, for direct access to rich LLM functionalities, such as page summarization and bilingual webpage translation.

🌟 Features

🔐 Sign in with ChatGPT:

Use your ChatGPT plan instead of an API key: choose Sign in with ChatGPT during onboarding or in Settings → AI provider. Chat, summaries, image actions, and translation then use the models your plan offers, with nothing to install.
Requests count toward your plan's usage, which you can review in ChatGPT settings.

💬 AI Chat Sidebar:

Quickly launch the sidebar using a keyboard shortcut (default: Ctrl/Cmd+Shift+E) to interact with LLMs via a side panel for queries, brainstorming, or information retrieval without navigating away from the current page.

🌐 Bilingual Reading:

View web content in two languages side-by-side, aiding in language learning or comprehension of foreign-language texts.

📺 LLM Translations for Youtube Captions:

Generate LLM based translations for Youtube captions in realtime.

📝 Memo & Note-taking:

Create, edit, and manage personal notes directly within the extension. Features auto-save, pinning important memos, and organized note management.

📄 One-Click Page Summaries:

Generate concise summaries of web pages for quick content overview.

🖱️ Context Menu Actions:

Right-click on any page to translate or summarize it instantly
Right-click on images to describe their content or extract text (OCR) using AI

🌍 Multi-language Support:

Available in 23 languages including English, Spanish, French, German, Japanese, Chinese, Korean, Arabic, Hindi, and more.

🤖 Multiple AI Model Support:

OpenAI: GPT-6.1 Sol, GPT-6 Luna
Google: Gemini 3.8 Flash, Gemini 3.5 Flash-Lite, Gemma 4 31B
Anthropic: Claude Sonnet 5, Claude Haiku 4.5
DeepSeek (OpenRouter only): DeepSeek V4 Pro, DeepSeek V4.1 Flash
OpenRouter: all of the above with a single API key
Local models: any model you run with Ollama, LM Studio, or llama.cpp, with no API key
ChatGPT plan: the models your plan offers, through Sign in with ChatGPT

🔑 Use Your Own API Keys:

Supports personal OpenAI/Gemini/Anthropic API keys for direct and the most cost-effective usage of models. Users are billed directly by those vendors.
Or use one OpenRouter API key for every model, billed by OpenRouter.

🎨 Color Themes:

Offers Light and Dark mode options for interface customization.

🛡️ Secure & Private:

API keys, ChatGPT sign-in, chat history, and memos are stored only in your browser. Page content is sent only to the AI provider you choose, when you use a feature.

🔍 Transparency & FAQ

Sign in with ChatGPT

Choose AI provider → Sign in with ChatGPT to use your ChatGPT plan with the existing chat, summary, image, and translation features. Click Continue with ChatGPT and finish OpenAI's sign-in in the tab that opens; there is nothing to install. Available models come from your signed-in account; usage and app limits are managed in ChatGPT settings.

Is it really free? Yes, and it will stay free. There is no Shizue account, paid plan, or ads. You can use your AI provider's API, a local model, or your ChatGPT plan. The Token Usage chart (the chart button below the chat box) shows how many tokens each model used; ChatGPT plan limits are managed in ChatGPT settings.

Is the whole thing open source? Yes. The full source of the extension is in this repository under the MIT License. The repository was private for a period after mid-2025 and is public again. It was AGPL-3.0 from June 2025 to October 2026 and is MIT as of v0.2.9. The PDF translation feature, which was built on AGPL-licensed BabelDOC, has been removed.

Does Shizue have a server? What leaves my browser? There is no Shizue server. API keys, chat history, memos, and settings stay in your browser's storage. When you use a feature, only the content that feature needs (e.g. the page text for a summary) is sent directly from your browser to the provider you chose, to OpenRouter, or to the local model server you set. There is no analytics or tracking. Details are in the privacy policy.

With Sign in with ChatGPT, the same content goes directly from your browser to OpenAI. OpenAI's sign-in returns to a 127.0.0.1 address, as its sign-in flow requires; nothing listens there, and Shizue catches that address inside the browser to finish signing in. OAuth tokens are kept in the extension's local storage, like API keys.

What does OpenRouter see about Shizue? Requests to OpenRouter carry HTTP-Referer: https://shizue.net and X-OpenRouter-Title: Shizue, so OpenRouter counts the usage under the Shizue app. These headers identify the app, not you, and nothing is sent to shizue.net.

Why these permissions?

storage: keep your keys, settings, and history locally
scripting and HTTP/HTTPS host access: reconnect Shizue in already-open web pages after an extension update or reload, without refreshing the pages
sidePanel: the chat side panel
activeTab: work with the tab you're on when you open Shizue or use a right-click action
contextMenus: the right-click actions
Content scripts on all sites: the floating toggle button and page translation; on YouTube, the caption translation
declarativeNetRequestWithHostAccess and access to localhost: reach a model server on your computer. Ollama rejects requests from browser extensions, so Shizue sets the Origin header of its own requests to Ollama to Ollama's address. During Sign in with ChatGPT, it also sends the browser's return to 127.0.0.1 to Shizue's own sign-in page. No other requests are changed.

Do existing pages need a refresh after an update? Shizue automatically reconnects its page scripts after extension updates, reloads, and browser restarts. It checks for an existing connection before injecting, preserves page contents and form input, and retries when a sleeping tab becomes active. Chrome settings pages, the Chrome Web Store, and sites where extension access is withheld cannot be reconnected this way.

Local models? Yes. In Settings → Models, choose "Use a Local Model". Shizue finds Ollama, LM Studio, or llama.cpp on their default ports, or you can enter an address. Pages and chats then stay on your computer. Translation quality depends on the model: in our tests, gemma4:12b translated a test page accurately, while qwen3.5:9b left some lines untranslated.

Firefox? Not yet.

🔀 Using Shizue with OpenRouter

One OpenRouter key gives you every model in Shizue, including DeepSeek, which is available only through OpenRouter.

Create a key at openrouter.ai/keys.
In Shizue's onboarding, choose Use OpenRouter API Key and paste it. Later you can change it in Settings → Models → AI Provider.
If you also register a provider's own key (OpenAI, Gemini, or Anthropic), its models use that key directly. Check Prefer OpenRouter to send them through OpenRouter instead.

🐳 Installation

You can install Shizue from Chrome Web Store,
or you can install Shizue manually from the source by following these steps:

Manual Install (for Developers):

Clone & Setup:

# Ensure pnpm is installed (npm i -g pnpm)
git clone https://github.com/rokrokss/shizue && cd shizue
pnpm install

Build:

pnpm build # For production build
# Or use `pnpm dev` for development with hot-reloading

Load in Chrome:
- Go to chrome://extensions.
- Enable "Developer mode".
- Click "Load unpacked" and select the build output directory (dist/chrome-mv3/).

💬 Community & Feedback

For suggestions, feedback, or discussions, join me on Discord.

📄 License

MIT