How to connect ChatGPT, Claude, and other AI assistants to your notes

You ask ChatGPT or Claude about work, health, money, and what to cook. Some answers are worth keeping. Most of them end up in a long list of chats sorted by date, and the next chat starts from zero.

A notes app you can connect fixes both sides of that. Your assistant can answer from what you already wrote down, and it can save the good parts of a chat back into your notes, where you can find them again.

This guide explains how that works, which assistants support it, and where it falls short. We make Capacities, so we use it for the examples. We also tell you when another app is the better choice.

The short answer

What "connecting your notes" actually means

When you connect a notes app to an assistant, the assistant gets a small set of tools it can use while it answers you. In Capacities, those tools are things like search the space, read this note, create a page, and save to today's note.

You keep chatting the way you always do. When your question needs your notes, the assistant calls a tool, gets back the notes that match, and uses them in its answer. When you say "save this", it calls another tool and writes a note.

Three things are worth knowing:

• It is not an upload. Your notes stay in your notes app. The assistant reads only what it looks up for the question in front of it.
• You choose the space. When you sign in, you pick one space. The assistant cannot reach any other space.
• It runs on an open standard. The technical name is MCP, the Model Context Protocol. ChatGPT, Claude, Le Chat, Perplexity, Cursor, and many other tools speak it. So one notes app can work with all of them, and your notes stay put if you switch assistants.

The two jobs: asking your notes, and saving to them

Asking your notes

This is the job most people search for. You have notes, and you want an assistant to answer from them instead of from the internet.

• "What did Jonas recommend for Lisbon?"
• "Summarize what I wrote about the renovation budget this year."
• "Which books have I finished that mention habits? Make a table."
• "I have a call with Maya at three. What did we decide last time?"

The answer is only as good as what the assistant can find. Notes with clear titles and types help a lot. A note called "Call with Maya" with Maya linked as a person is easy to find. A page called "Untitled 14" is not.

Saving to your notes

This is the job most people do not know they can do. At the end of a good chat, you tell the assistant to keep the part that matters.

A short note like this is easier to find and reuse than a whole chat transcript. Our guide to saving AI chats as notes shows prompts that work and a weekly routine.

Which assistants can connect

Capacities works with any assistant that supports MCP. These are the most common, checked in October 2026. Plan rules on the assistant's side change often, so check there before you buy.

The setup guide has step-by-step instructions for each of these, plus Raycast, ClickUp, Langdock, Manus, Dust, and others.

What the assistant can do, and what it cannot

Here is what a connected assistant can do in a Capacities space today:

• Search by keywords and topics, and list notes by type, tag, or collection.
• Read whole notes, including their properties, such as a book's author or a meeting's date.
• Create pages, tasks, and objects of any type you have, such as a Book, a Person, or a Recipe, with their properties filled in.
• Edit existing notes: change properties, add to the end, and add, rewrite, or delete paragraphs.
• Save text to today's daily note.
• Upload images and PDFs you share in the chat.
• Link to notes in its replies, so you can open them in Capacities.

And here is what it cannot do yet. We would rather you know before you pay.

• It cannot change the structure of your space. It can fill in the properties a type already has, but it cannot create a new object type or add a property to one. You do that in Capacities.
• It is weak at "read everything". A prompt like "go through all 400 of my notes and count the flights" works poorly. Ask it to search or list first, such as "list my Trip objects from this year", then read those.
• It works on one space per connection. If you keep work and personal notes in two spaces, you add two connections.

Built-in AI memory is not the same thing

ChatGPT and Claude both have memory, and both have Projects. They are useful, and they solve a different problem.

• Memory keeps facts about you, chosen by the AI. OpenAI's help page says memory "does not retain every detail from every conversation. ChatGPT decides which available information is relevant." Claude's memory is a summary of your conversations, updated every 24 hours, focused on things like your role and preferences.
• You cannot open it like a notebook. There is no page for "what I learned about index funds" that you can read, edit, and link to your Money page.
• It stays inside one app. If you use ChatGPT at home and Claude at work, each one remembers a different half of you.
• Projects hold files for one effort. A ChatGPT project holds 5 files on Free and 25 on Plus. That is good for a single project, and it is not a place for everything you know.

Your notes are the opposite. You decide what goes in, every note is a page you can open, and every connected assistant reads the same notes. Use memory for your preferences. Use notes for what you know.

Why the shape of your notes matters to an AI

An assistant can only use what it can find and understand. Three things make a real difference.

Types tell the AI what a note is. In Capacities, a note can be a Book, a Person, a Meeting, or any type you create, and each type has its own properties. When you ask for "the books I finished this year", the assistant lists your Book objects and checks the status. In a folder of loose pages, it has to guess from titles.

Links give it context. When a meeting note links to Maya and to the launch plan, the assistant can follow those links. It learns who Maya is and what the launch is without you explaining it again.

A daily note gives it somewhere to put things. "Save this to today" always works, even when neither of you knows yet where the note belongs. You sort it out later, or never.

Privacy and control

• One space per connection. The assistant reaches only the space you choose when you sign in.
• You can see and remove every connection. They are listed under Settings > AI Assistant in Capacities. Removing a connection in Capacities does not delete it in the AI app, so remove it there too.
• Version history on every note. If an assistant changes something you did not want, restore an earlier version.
• The AI app's rules apply to what it reads. ChatGPT and Claude have their own data settings, including whether chats are used to train models. Check those settings.
• Your notes are stored in the EU. Capacities is made by Capacities Labs GmbH in Freiburg, Germany.

What it costs

Connecting an assistant is part of Capacities Pro. You can build your notes on the free plan first and add Pro when you want to connect.

The AI itself is paid for by the app you chat in. Chats in ChatGPT or Claude do not use your Capacities AI budget. If you already pay for ChatGPT Plus or Claude Pro, you do not pay for that AI twice.

Guides for each assistant

• Connect ChatGPT to your notes: for people who use ChatGPT every day and want it to remember what matters.
• Connect Claude to your notes: for the Claude app, Cowork, and Claude Code.
• Build an AI second brain for your agents: for Cursor, Claude Code, Codex, and other agents.
• Save AI chats as notes you will find again: the routine, with prompts you can copy.

Who should not use this

• If you want an agent to work on plain files on your disk. Coding agents such as Claude Code can read and rewrite a folder of Markdown files directly. If that is how you work, Obsidian or Logseq fit better. Capacities exports to Markdown, but it is not a file-based app.
• If your team needs shared AI search. Capacities is for one person. A team wiki such as Notion, Confluence, or Slite fits a shared knowledge base better.
• If you do not want AI near your notes at all. That is fine. Capacities works without AI, and you can hide AI in the app.

How we wrote this guide

• We make Capacities. The examples use it, and the limits section is ours to keep honest.
• Facts come from official sources, read in October 2026: our own docs, OpenAI's and Anthropic's help centers, and each assistant's connector documentation. Plan rules on the assistant's side change often.
• The memory comparison quotes the vendors' own help pages. We link them so you can check.
• If something here is wrong or out of date, email [email protected] and we will fix it.