We’ve all felt the broken workflow of AI chatbots: you start reading a long response → hit a confusing sentence → ask in the composer → jumps to the bottom and appends more text, pulling you away from the exact context that triggered the question. After about 5 rounds of back-and-forth, we typically lose track of the original knowledge structure, making it difficult to build on earlier points in the conversation.
Coo is a light-weight AI helper for Obsidian, which makes the discussion happens at the point of friction, not at the end of the thread.
https://github.com/jwy600/obsidian-coo
With coo, you can:
Ask: type any question, and the AI response is saved as a new note in your vault.
Discuss: Select text in your editor and open the Discuss composer from command palette or contextual menu. You can type a custom question, or choose from quick action buttons.
Annotate: When you find something worth keeping, you can select text in the textarea and save it as a piece of annotation attached to the paragraph.
Rewrite: Place your cursor in a paragraph that has annotations below it and run coo: Rewrite from either command palette or contextual menu. The AI rewrites the paragraph incorporating your picked phrases — translation of a key term, a concrete example, a simpler explanation — then replaces the original text and removes the annotations. Fully undoable with Ctrl/Cmd+Z.
Inspire: Write an {instruction} anywhere in a paragraph — for example, {give 3 examples} — and run coo: Inspire.The AI generates 2–5 concise bullet points based on your instruction and appends them directly after the paragraph. If the paragraph is a list item, bullets are automatically nested. The {instruction} is removed from the text after expansion.
Chatbots create a kind of “fourth wall”. Knowledge performs in front of you—confident, linear, self-contained—while you sit in the audience. When a sentence is unclear or a paragraph doesn’t match what you need, you can’t work where the problem lives. You have to finish the show, then prompt again and hope the next performance lands closer.
The plug-in is made to improve this UX. Happy to learn feedback.