Marginalia: Margin notes for AI chats 作者: Midhun
Google-Docs-style follow-up comment threads on Gemini, ChatGPT and Claude — highlight text, ask in the margin, keep your place.
擴充套件後設資料
畫面擷圖
關於此擴充套件
Google-Docs-style follow-up comment threads on Gemini, ChatGPT and Claude — highlight text, ask in the margin, keep your place.
Ask follow-up questions in the margins - without burying the answer you're reading.
When an AI gives you a long answer and one phrase needs unpacking, you normally ask in the main chat and scroll away from your place. Marginalia gives you Google-doc-style comment threads instead: highlight the phrase, ask your question in a comment box that opens right next to it, and the reply streams into the box while the original answer stays put.
How it works:
* Highlight text in an AI answer on chatgpt.com or claude.ai or gemini.google.com
* Click the Comment pill ( or right-click -> "Ask About..." or press Ctrl + Shift + H)
* A comment thread opens in the right margin, anchored to your highlight
* Ask follow-ups in the thread - replies come from the AI you're already using.
Features:
* Multi-turn threads that persist per conversation and re-anchor when you return
* Docs-style layout: focusing a thread aligns it with its highlight
* Minimize threads to chips, expand to full screen, or resolve them for later
* An all-threads panel (Alt + Shit + A) with open/resolved filter and click-to-jump
* Follows each site's light/dark theme
* Keyboard-first: cycle threads with Alt+↑/↓, re-bindable shortcut
Your data stays yours:
Threads and settings are stored locally on your browser. Your questions go only to the AI service you are already using - either through your existing signed-in session or through the providers official API if you add your own API key in Settings ( ChatGPT requires an OpenAI key). Keys are stored locally and sent only to that provider. Marginalia has no server and collects nothing.
The source code is open-source and is available for community review.
Ask follow-up questions in the margins - without burying the answer you're reading.
When an AI gives you a long answer and one phrase needs unpacking, you normally ask in the main chat and scroll away from your place. Marginalia gives you Google-doc-style comment threads instead: highlight the phrase, ask your question in a comment box that opens right next to it, and the reply streams into the box while the original answer stays put.
How it works:
* Highlight text in an AI answer on chatgpt.com or claude.ai or gemini.google.com
* Click the Comment pill ( or right-click -> "Ask About..." or press Ctrl + Shift + H)
* A comment thread opens in the right margin, anchored to your highlight
* Ask follow-ups in the thread - replies come from the AI you're already using.
Features:
* Multi-turn threads that persist per conversation and re-anchor when you return
* Docs-style layout: focusing a thread aligns it with its highlight
* Minimize threads to chips, expand to full screen, or resolve them for later
* An all-threads panel (Alt + Shit + A) with open/resolved filter and click-to-jump
* Follows each site's light/dark theme
* Keyboard-first: cycle threads with Alt+↑/↓, re-bindable shortcut
Your data stays yours:
Threads and settings are stored locally on your browser. Your questions go only to the AI service you are already using - either through your existing signed-in session or through the providers official API if you add your own API key in Settings ( ChatGPT requires an OpenAI key). Keys are stored locally and sent only to that provider. Marginalia has no server and collects nothing.
The source code is open-source and is available for community review.
由 1 位評論者給出 0 分
權限與資料
必要權限:
- 存取您在 gemini.google.com 的資料
- 存取您在 chatgpt.com 的資料
- 存取您在 chat.openai.com 的資料
- 存取您在 claude.ai 的資料
選用的權限:
- 存取您在 gemini.google.com 的資料
- 存取您在 chatgpt.com 的資料
- 存取您在 chat.openai.com 的資料
- 存取您在 claude.ai 的資料
- 存取您在 api.openai.com 的資料
- 存取您在 generativelanguage.googleapis.com 的資料
- 存取您在 api.anthropic.com 的資料
開發者聲稱必須收集下列資料:
- 網站內容
更多資訊
- 版本
- 0.2.2
- 大小
- 177.86 KB
- 最近更新
- 3 天前 (2026年7月23日)
- 相關分類
- 授權條款
- MIT License
- 隱私權保護政策
- 閱讀此附加元件的隱私權保護政策
- 版本紀錄
- 新增至收藏集