When an AI helps you write, you want the paragraph you are working on to rest only on the sources you cited there. When you are working out a structure or what to try next, you want it to draw on all your notes so far. It is the same AI, but what you want it to look at changes with the moment.
I wanted that switch to happen without spelling it out in every request. So the chat beside a note has a Grounding chip by the input box with three choices. The default is This note: the AI gets the open note and what you cited in it with @, and nothing else. It does not go looking through your other notes. When you are working out a structure or looking for ideas, switch to Internal, and it also searches across your notes and knowledge. External adds a web search on top of that (this needs a way to search the web, such as a search MCP server).
For a conversation that is not about any one note, start it from Chat in the left navigation. It always searches across your notes and knowledge, so it offers two choices, Internal and External. You can also narrow it to one folder with Folder to search.
There are limits. What you narrow is the material handed to the AI; it does not shut out the general knowledge the AI already has. In the chat without a note, you cannot attach notes with @ yet.
The manual's page "Chat & Ask" covers how the choices differ. Choosing the grounding arrived in v0.16.8 (2026-07-02), the chat without a note in v0.81.0 (2026-09-18), and the folder choice in v0.110.0 (2026-11-02).
https://kumagallium.github.io/Graphium/manual/ai-chat-and-ask
Graphium is on GitHub.

Top comments (0)