English | 中文
Select any sentence on a web page and get an explanation based on your LLM's understanding of the entire article — not just the snippet. Then ask follow‑up questions, in any language, with optional live web search and cited sources.
Most "explain / translate selection" extensions send only the highlighted snippet to the model, so the explanation is out of context. InContext extracts the whole article first (via Mozilla Readability) and feeds it to the model as grounding, so it explains what the selection means in this specific article — the author's intent, implied meaning, what a term refers to here.
| Selection translators | Full‑page summarizers | InContext | |
|---|---|---|---|
| Explains the selected passage | ✅ | ❌ | ✅ |
| Uses the whole article as context | ❌ | ✅ | ✅ |
| Explains the selection using the whole article | ❌ | ❌ | ✅ |
| Multi‑turn follow‑up on that passage | partial | ❌ | ✅ |
| Live web search with cited sources | ❌ | ❌ | ✅ |
| Bring your own key, 15+ providers | varies | varies | ✅ |
One line: Not "translate this sentence" — "tell me what this sentence actually means in this article."
- Whole‑page grounding — the entire article is the context, not the isolated snippet.
- Three‑step explanations — term definition → meaning in context → example when helpful.
- Multi‑turn chat — keep asking about the same passage.
- Any output language — Chinese / English / 日本語 / …
- Web search (hybrid) — optionally pull live facts: a search engine fetches data, your main model writes the answer and cites sources with [n] superscripts. Context‑aware query generation (bilingual, disambiguates homonyms like ReAct vs React).
- 15+ model providers, BYOK — DeepSeek, GLM, Kimi, Qwen, Doubao, OpenAI, Gemini, Claude (native), Grok, Mistral, Groq, OpenRouter, SiliconFlow… Keys stored locally, never sent to any third party. Each provider's key is saved separately.
- Per‑tab — each tab keeps its own conversation; switching tabs never overwrites or interrupts a running explanation.
➕ Add to Chrome — one click, then open the side panel → ⚙ → choose a provider and paste your API key.
- Download / clone this repo.
- Open
chrome://extensions, turn on Developer mode. - Click Load unpacked, pick the project folder (the one with
manifest.json). - Pin the icon. Open the side panel → ⚙ → choose a provider and paste your API key.
Select text on any page → click the floating Explain / 解释 button (or right‑click menu / Ctrl+Shift+E) → the side panel explains it using the whole article → ask follow‑ups below. Toggle 🔍 Web for live, cited answers.
Try it on: Lilian Weng — LLM Powered Autonomous Agents — highlight ReAct and watch it explain the LLM pattern (not the React framework).
Content script detect selection → floating button → Readability extracts full article → send to background
│
Background store selection per‑tab/window → open Side Panel → notify
│
Side Panel (optional) generate search query → web search → build prompt(full article + selection + sources)
→ stream from your model → multi‑turn UI with [n] citations
Core grounding prompt: src/prompt.js. Web search: src/search.js. Providers / Anthropic adapter: src/llm.js & src/config.js. No build step — plain MV3 + vanilla JS.
Is my data private? Yes. It's BYOK — your API key and conversations stay in your browser's local storage. The page text goes only to the model endpoint you configured.
Which models can I use? Any OpenAI‑compatible endpoint plus native Anthropic Claude — 15+ presets, or add a custom one.
Does it work on PDFs / local files? Local file:// pages need "Allow access to file URLs" enabled for the extension; some Chrome internal pages can't be accessed at all.
MIT
English | 中文
划选网页中的任意一段，基于大模型对【整篇文章】的理解给出解释——不是孤立翻译那一句。还能就这段多轮追问、多语种输出、联网补充并标注来源。
普通"划词翻译/解释"插件只把选中的那一小段发给模型，解释脱离语境。InContext 先用 Readability 抽取整篇文章喂给模型做背景，所以它解释的是这段话在这篇文章里到底什么意思——作者意图、隐含信息、专有名词在此处的真实所指。
| 划词翻译 | 全文总结 | InContext | |
|---|---|---|---|
| 解释选中段 | ✅ | ❌ | ✅ |
| 拿整篇文章当上下文 | ❌ | ✅ | ✅ |
| 用整篇文章来解释选中段 | ❌ | ❌ | ✅ |
| 就该段多轮追问 | 部分 | ❌ | ✅ |
| 联网搜索 + 标注来源 | ❌ | ❌ | ✅ |
| 自带 key、15+ 服务商 | 不一定 | 不一定 | ✅ |
一句话： 不是"翻译这句话"，而是"告诉我这句话在这篇文章里到底什么意思"。
- 基于全文：整篇文章是上下文，而非孤立片段。
- 三步式解释：术语释义 → 全文语境 → 必要时举例。
- 多轮追问：就同一段持续提问。
- 多语种输出：中文 / English / 日本語 …
- 联网（混合模式）：可选实时联网——搜索引擎取数据，你的主模型作答并用 [n] 上标标注来源。检索词结合上下文生成、中英双语、自动消歧（如 ReAct ≠ React）。
- 15+ 模型，自带 key：DeepSeek、GLM、Kimi、通义、豆包、OpenAI、Gemini、Claude（原生直连）、Grok、Mistral、Groq、OpenRouter、硅基流动… key 只存本地、不传第三方，每家各存各的。
- 按标签页独立：每个 tab 各记各的对话，切换不覆盖、不打断正在跑的解释。
Chrome 应用商店（推荐）： ➕ 一键添加到 Chrome —— 装好后开侧栏 → ⚙ → 选服务商、填你的 API Key。
开发者模式加载（备选）：
- 下载/克隆本仓库。
- 打开
chrome://extensions，开启开发者模式。 - 点加载已解压的扩展程序，选含
manifest.json的文件夹。 - 固定图标。开侧栏 → ⚙ → 选服务商、填你的 API Key。
任意网页划选文字 → 点浮出的「解释」按钮（或右键菜单 / Ctrl+Shift+E）→ 侧栏结合全文解释 → 在下方继续追问。点亮「🔍 联网」获取实时、带来源的回答。
试用建议： 打开 Lilian Weng《LLM Powered Autonomous Agents》，划选 ReAct，看它解释成大模型的"推理+行动"范式（而不是前端 React）。
隐私安全吗？ 安全。BYOK——你的 key 和对话只存在浏览器本地；网页正文只发往你自己配置的模型接口，无任何后端。
支持哪些模型？ 任意 OpenAI 兼容接口 + 原生 Anthropic Claude，15+ 预设，也可自定义。
支持 PDF / 本地文件吗？ 本地 file:// 页面需给扩展开启"允许访问文件网址"；部分 Chrome 内置页面无法访问。
MIT