Hey folks, I’m looking for an Obsidian plugin or setup that gives me a simple sidebar chat interface — something where I can:
Ask questions about the currently selected note or selected text
Summarize or modify the note
Occasionally generate follow-ups, new notes, etc.
Basically, I want something like Cursor AI (from dev/IDE world) but for my Obsidian notes.
Important:
I don’t want to use an OpenAI API key or pay extra per token. I want to connect my own ChatGPT or Claude account, ideally via browser/session or plugin.
I’ve already tried:
Smart Connections
Copilot for Obsidian
ChatGPT MD
Unfortunately, none of them fully support this workflow — they either don’t reference the selected note well, don’t support browser-based accounts, or lack a clean sidebar flow.
Has anyone found a plugin or workaround for this? Even better if it works with Claude or ChatGPT web accounts directly.
What about your notes do you need ask questions about? A little confused on the usage here. It seems you know how to use the LLM’s, so I guess my recommendation would be just copy / paste between obsidian and your chats in a browser.
Just did some looking into the plugins in the marketplace and the at least the ones I looked at all use API keys, I don’t know if a random plugin dev would be successful in getting an integration setup with one of the LLM services.
It doesn’t, because it’s a question asking if there’s a plugin they’ve missed, or some other way to accomplish the goal. But if there isn’t a plugin, much of the text could certainly be used in a Plugin Ideas post.
When creating your prompt, you can use several special placeholders:
{}: Represents the selected text in your note.
{[[Note Title]]}: Represents the content of a specific note.
{activeNote}: Represents the content of the currently active note.
{FolderPath}: Represents all notes within a specific folder.
{#tag1, #tag2}: Represents all notes with any of the specified tags. Note that the tag must be in the note property to work!
For the selected text and notes, whether current (active) or added by wikilinks, no vectors need added, for the others ostensibly you’ll need to vectorize vault with an embedding model.
As for CursorAI like experience, I’ve not seen anything like that automatism.
I seem to remember a post where similar effort was shown (something to do with proofreading and maybe only OpenAI was added anyway), but I cannot find it. It may have been @SkepticMystic’s effort but I cannot see the repo now on his GitHub. Maybe it was someone else.