Use caseWrite
Ask your AI app about your own library
A two-page brief in your folder, written in the AI app you already work in, quoting your own papers — and a set of changes to it you accepted one by one.
- You start with
- A folder of papers and books you have not finished reading, and Claude Code, Codex or Claude already open for something else.
- Time
- About an hour, most of it spent reading what comes back.
- Your part
- The question, the judgement on what the brief is missing, and every edit you accept.
What you'll end up with
- A brief written into the folder, with quotations that open on the page they came from.
- A long book answered from the chapters that matter, not from a search result.
- A short list of changes the app proposed to the brief — accepted, rejected or rewritten by you.
- The whole session listed in MindMax afterwards, call by call, so you can see what it actually read.
The point is not that an agent can read a file. It is that it read the shelf, and that you can check what it says.
How to hand your library to an AI app, step by step
Open the shelf as a workspace
YouPoint MindMax at the folder holding the papers and books, and leave it in front. That open workspace is what a connected app can reach.
Connect the app you already use
YouSettings → External AI, then Connect. MindMax writes the app’s configuration, offers a restart if it needs one, and lists the app as connected once it makes a real call.
Ask a question that spans the shelf
You ask · MindMax answersAskUsing my MindMax library, what do these papers actually agree on about the effect, and where do they disagree?
The app searches the whole workspace, not one file. Each hit comes back with its section and a handle it can cite, so the answer starts from passages it can point at.
Send it into the book that matters
AskRead the third chapter of the book I have open in MindMax and tell me what it argues.
AskGo deeper on the section about the trial design.
A long book arrives as a map — chapters, one-line notes, section ids. The app chooses a chapter, reads it whole, and comes back for another if the question is not settled.
Ask for the brief, in the folder
You ask · MindMax answersAskWrite a two-page brief for my team from these sources, with a quote for every claim, as a new document in this folder.
The file is created through MindMax, so it lands where you asked, appears in the file tree and is recorded in history under the app’s name.
Review the changes it proposes
YouAsk it to tighten the argument. The rewrite arrives as a proposal in MindMax — grouped tracked changes with reasons, on your text. Accept the groups you agree with, reject the rest.
Export it as the file people expect
AskExport that brief to Word next to the original.
The export goes through MindMax’s own exporter, so tables stay tables and numbered sources stay numbered.
Why it works here
A book is navigated, not skimmed
The agent does not get four thousand pages of text and a hope. It gets a map with a note per chapter and a section id for each, then reads whole sections — the same route MindMax’s own assistant takes.
How the connection works →Quotes are expanded, not trusted
A search hit carries a handle. When the brief is saved, MindMax expands that handle into the verbatim passage with a clickable anchor — so a citation is a place in a document, not a line the model typed from memory.
Why checked quotes matter →The writing comes back for review
New files are created outright; changes to something you wrote arrive as tracked changes you accept group by group. There is no overwrite and no delete in the tool set.
Edits wait for you →Also used here: Guide: connect an AI app · All the tools it can call · Thinking methods on a canvas · Export to Word and PDF
Where it gets hard
- You pay the app, not MindMax
The connection spends no MindMax credits because no MindMax model runs on it. The reading and writing are tokens in your Claude or OpenAI account, and a long book costs roughly what reading it once costs.
- Which app you use matters
Claude Code and Codex run long jobs comfortably and can spread parallel steps across sub-agents. A chat app with a small context does the same work in smaller packages, more slowly — the tool says so before a long method starts.
- It cannot see what you never opened
The library is the workspace you opened in MindMax. Files elsewhere on the disk are not in scope, absolute paths are refused for writing, and a folder of scans without recognition reads as images.
- A brief is still a draft
Citations make claims checkable; they do not make them correct. The app can attach a real quote to a wrong conclusion. Read the brief, open two or three citations, and change what does not hold.
- Some app builds are unverified
Claude, Claude Code and Codex are verified end to end. Cursor, Windsurf and WorkBuddy are set up for you and confirmed on first use; the ChatGPT desktop app may not surface local tools in a given build.
Then what
About 30 minutes
When the library is one paper you have to present tomorrow.
An afternoon
When the brief is the start of something longer and the sources need organising.
Ten minutes
When you do not yet know what is in the library you are about to hand to an agent.
The connection is free; a readable library is the part that takes work.