Carrel — your whole term, and everything you wrote about it

You wrote it down. Finding it is the hard part.

It comes four months later, when a professor's deadline moves, or a final paper needs the reading you annotated in September — and the answer is in a notebook, a thread, a PDF, a recording, or a message to yourself. Somewhere. Notes apps store text you have to remember you wrote. Chat assistants forget you between sessions. Neither can answer what did I say about this after the fact.

Open Brain treats that as a retrieval problem rather than a filing problem.

You open Carrel. Open Brain does the remembering.

Everything you put in — a thought, a PDF, a link, a recorded lecture, a syllabus — goes into Open Brain: one memory, read and understood, in a database that belongs to you. You never have to remember which folder or which week.

Carrel is the part you actually open. Your day, your week, your term, and everything you've ever captured. A searchable archive of your whole degree isn't something a person can use on a Tuesday morning — a page that says what's due, what's late, and what you read about it in September is. That second half is the entire reason this is a product and not a database.

Nothing needs filing, so nothing gets lost

Coursework, and everything that isn't: research, writing, side projects, a job search — the things with no syllabus and no due date, where what you know is spread across your own notes.

Capture a thought the way you'd text yourself.
Drop a PDF, a Word file, an image, a whole folder. Paste a link and the page itself is read, not just its address.
Upload a lecture recording and it comes back as text — searchable by what was said, beside the notes you took in the room.
Papers, articles and saved links end up in the same place as the notes you wrote about them.
If you write code, every commit goes in too — in every repository on the machine, without you doing anything — so months later you can search why you changed something, not just when.

In through whichever door is nearest.

The Library screen's Capture tab: a box for typing a thought, and buttons for reading a link, adding files, and adding a whole folder — above a note that anything dropped on the page, from PDFs and slides to images, audio and video, is read and made searchable. — Every door in one place: type a thought, paste a link, add files, or point it at a folder. The size limits are stated before the upload rather than after it, so a recording too long to transcribe says so while you can still do something about it.

See the weeks that will hurt throughout your term

Drop one in and every deadline is proposed — for you to accept or reject. Nothing is written until you say so.

The week that will hurt is visible in September, while there's still time to move work out of it. What you get back is the term as a load chart — every week of it, drawn before any of them arrive.

Re-upload a revised syllabus and a moved midterm is offered as a reschedule, not a second deadline sitting next to the old one.

Every course, deadline and bar in the picture was written by the software itself — six syllabi went in through the upload box above. The term is invented; the reading of it is not.

The Plan screen for Fall 2026: twelve weekly bars of committed load with exam markers beneath them, the current week carrying a red cap for three overdue tasks, a line reading 'Heaviest week is Sep 21 — 27 against a typical 22', and a written assessment naming the weeks to watch. — Fall 2026, drawn from six syllabus files. The tallest bar is the week to move work out of. The red cap is work already overdue — it sits above the bar rather than inside it, so a week's height is how full it is, not how far behind you are.

Know whether Friday afternoon is free

Your day and your week ordered by what's actually owed, not by what you last typed.

You can see that Friday afternoon is free without working it out. Every class is a block as long as the class actually runs, so a full day looks full. What's late, what's due today and what's planned sit underneath, and anything there can be ticked off where it sits.

The home screen on a Wednesday: three overdue and three due today, a seven-column week with each class drawn as a block at the hour it runs and for as long as it lasts, and a Today list of three late assignments each with a checkbox. — A Wednesday: three overdue, three due today, and every class drawn as long as it runs. Nothing here is a start time — a 50-minute class and a 90-minute lab are different heights, which is what makes an empty afternoon visible.

Why page 40 is findable, not just the title

Six ways in, one memory, two ways out — you, or whatever AI you're already talking to. The part worth explaining is the middle, because it's the part you never see.

Text is extracted, summarized and indexed by meaning. Images are described, so both the text in them and what they show are findable. And a long document is broken into passages, each one indexed with a sentence saying where it sits — which is what makes a paragraph on page 40 retrievable rather than just the title.

One long document48,320 characters

Too long to be a to-doPasted in, it is kept as the document it is — and the one commitment buried in it still becomes a task.

Summarized so it can be foundThe first line names the people in it and the date, because that is how you go looking for something months later.

Split into passages29 passagesEach stored with a sentence situating it in the document, so “he said that would be better” is searchable.

AnswerableAsk a question and the passage comes back — with the sources it used, so the answer is checkable rather than taken on trust.

Get the paragraph, not forty pages

Search by describing the thing. Ask what a reading said about something in September and get the passage — not a link to forty pages.

Most search is good at finding a document and bad at finding the sentence, so this gets tested rather than assumed. The benchmark: one real 48,320-character document, and six questions it genuinely answers — each answer buried between a third and four-fifths of the way in, never near the top.

The first bar is what you get from most tools: nothing, stated confidently. The last two answers were the interesting ones — they had been found, and were still buried too far down the results to reach you. Finding something and putting it in front of you are two different jobs, and the answer you get back does both.

The document is a real two-hour transcript out of the corpus — the same one the section above walks through. Nothing measured here is specific to a transcript; it is the length, and how deep the answers sit, that make it a test.

0/6Nothing

4/6Found but buried

6/6In front of you

It works with the AI you already use

Open Brain isn't another chat window you have to remember to open. It speaks MCP, the emerging standard for giving AI assistants access to your own data — so the assistant you already work in can search your memory, and write back to it, without you copying anything across. You stop re-explaining yourself every session.

Which means the memory outlives the tool. When you change assistants next year, what you've accumulated stays where it is.

$ claude mcp add --transport http open-brain
    https://your-project.supabase.co/functions/v1/open-brain-mcp

Added. 31 tools available:

search_thoughtscapture_thoughtsearch_moments log_momentsearch_contactscreate_task list_taskslist_eventssearch_materials get_materialsearch_commits+20 more

Where your information actually lives

The first fair question about any tool like this is where the information goes and how you'd get it back. Plainly:

One database, and it's yours. Not a shared index, not a folder inside somebody's product — an ordinary database you can be handed the keys to, in a format anything can read.
Everything comes back out. It's ordinary rows and your original files. You can walk away with all of it, and nobody has to approve it.
Read-only where it touches your other tools. Calendar is imported and never written back — and if you hide those events, the calendar tells you they're hidden instead of just looking empty. Nothing Open Brain does can change the systems it reads from.
Your own text can't hijack the assistant. Anything you wrote is handed to a model as something to read, never as something to do — a real failure mode in tools that skip this.
Credentials never live in the memory. The brain holds no secrets about itself, and it isn't where passwords or regulated records belong.

Your whole term, and everything you wrote about it.