Skip to main content

Use caseRead

Run a long job with your own AI agent

A whole book translated, a scan turned into searchable text, or a chapter-by-chapter reading guide — produced by the model you already pay for, saved in the same place MindMax would have saved it.

You start with
One long document in the folder MindMax has open, and a connected AI app with a model you have already paid for.
Time
An afternoon to set up and check the first steps; the rest runs while you do something else.
Your part
Choosing the method, reading the first two steps, and deciding whether the rest is worth the tokens.

What you'll end up with

  • The same deliverable a MindMax run would produce: a side-by-side translation, recognised pages with a per-page quality record, or margin notes in the reading panel.
  • A step-by-step record in the app’s session entry — which steps finished, which were returned for a fix, which are still open.
  • Quotes and chapter notes saved into your library, so later questions can find them instead of re-reading the book.
  • A resume point. Close the app, come back tomorrow, and press Continue — completed steps are not redone.

MindMax still owns the plan and the checks. Only the thinking moves to your app’s model.

How to run a MindMax method with your own model, step by step

  1. Pick the job and the material

    You

    A book to translate, a scan with no text layer, or a long report that needs a study guide. Open that folder as the workspace in MindMax.

  2. Connect the app whose model you want

    You

    Settings → External AI → Connect, with Collaborate access. Reading is free to MindMax on this path; the tokens are billed to that app.

  3. Name the method in plain words

    You ask · MindMax answers

    AskUse your own model to run a study guide for the book I have open in MindMax, with the critical lens, and save it where MindMax would save its own.

    You name the job, never the method id. MindMax finds the method, cuts the material into steps and returns the first work package.

  4. It works one step at a time

    Your AI app

    The app asks for a step, does that step with its own model, and submits it. MindMax validates the output, anchors quotes and saves it before naming the next steps.

  5. Read the first two steps

    You

    This is the cheapest place to catch a wrong lens, a bad terminology choice or a page that needs to be looked at as an image rather than read as text.

  6. Parallel steps go to sub-agents

    Your AI app

    Steps marked parallel can be worked on at once. A result that fails validation comes back with the reason and can be resubmitted twice.

  7. Open the result in MindMax

    You ask · MindMax answers

    AskOpen the finished result in MindMax so I can read it beside the source.

    A translation opens side by side, margin notes in the reading panel, chapter notes in the book’s map. If the app dropped out midway, press Continue first.

Why it works here

The model can be the one you already chose

You may have a subscription, a stronger model, or a local one. MindMax does not substitute a model of its own on this path; the app brings its model and its bill.

Bring your own key →

The plan and the validators stay with the library

Chunking, quote expansion and per-step checks are MindMax’s, so the result lands where its own run would have put it and stays usable by later questions.

MindMax Agent →

Interrupted work keeps its place

Finished steps are saved as they are validated. Resume from the same point instead of starting the book again — the same shape as a job MindMax runs itself.

Handing the assistant a long job →

Also used here: Guide: long jobs with an agent · Guide: connect an AI app · Every method an app can run · Translate a book inside MindMax · Recognising scanned PDFs

Where it gets hard

  • The tokens are real

    A full pass costs roughly one read of the material in the app’s context. No MindMax credits are spent, but that is not the same as free — ask the app to tell you before starting a long book.

  • A weak model stays a weak model

    MindMax supplies the plan, the map and the validators, not the reasoning. A model that misreads a technical chapter will still misread it, one validated step at a time.

  • Scans depend on the scan

    OCR quality varies by page. MindMax reports which pages came out badly instead of quoting them as clean text; look at those pages as images before relying on them.

  • The app must stay connected

    If the app disconnects, progress is kept but the run stops advancing. Press Continue in MindMax or ask the app to pick the run up again.

  • This is not a one-click book

    Long jobs are many steps, and you should read the first two. If you want MindMax to own the whole run without an app in the loop, its own assistant does that on your credits.

Cite this page