Docs / The ModelBrain menu
ModelBrain's icon lives in your menu bar on macOS and in your system tray on Windows. One click opens a menu that tells you whether ModelBrain is working and gives you the controls you will use most: add a folder, pause it, connect ChatGPT, turn reasoning on, restart. This page lists every item, top to bottom.
On macOS it sits in the menu bar at the top right of your screen. On Windows it sits in the system tray at the bottom right; if you do not see it, click the small up arrow next to the clock to show hidden icons. If the icon is missing on either system, ModelBrain is not running: open it from Applications (macOS) or the Start menu (Windows).
Items that only show information, like the status line, are not clickable. Checkbox items show a tick when they are on.
The first three rows answer one question: is ModelBrain working, and if not, what do I send for help?
A live check of the memory server. Running means your assistants can remember and recall. Not responding means the server is not answering: try Restart ModelBrain from this menu first.
Copies a status report to your clipboard in JSON form. Paste it into a message to support and it carries the details needed to see what is going on. It works even when the memory server is not responding, which is exactly when you need it.
Opens a small window with two parts: a Status table (whether the memory server is reachable, where your vault is stored on disk, and a row for each assistant) and a card for each assistant that has a one-click setup button. See Claude Desktop, Claude Code and Cursor.
From 0.6. Appears when an assistant has forgotten memories in the last 15 minutes, with the count in brackets. Choosing it brings them back as new copies, so a restored memory has a new key. Anything older than 15 minutes cannot be undone from here. To stop a runaway assistant, ModelBrain also refuses more than 20 forgotten or archived objects a minute from one assistant, counted across reconnects (a forget of a chain of 5 versions uses 5), and tells the assistant to wait. A forget that would remove more than that in one call is refused outright. Set MODELBRAIN_DESTRUCTIVE_PER_MINUTE to change the limit, or 0 to turn it off.
From 0.6. The first time an assistant ModelBrain has not seen before connects, it is served and a small window asks you to Allow or Block it. Closing the window leaves it undecided: it keeps working, and you are asked again after the menu helper restarts. A blocked assistant is refused with a client_blocked error. Assistants that come with ModelBrain itself are not asked about, except the Claude Code prompt hook, which is asked about once as modelbrain_claude_code_hook.
Opens the Account window. Without an account it offers Sign in or Add this computer (six matching words on both screens). Signed in, it shows your plan and renewal date, this computer's sync state with a Sync now button, your backups with Back up now and Restore, Export, and the apps connected through the relay with a Disconnect button for each (a newly connected app shows Allow and Deny until you decide). Manage billing, Cancel subscription and Resume subscription are here too. Everything is explained in Your ModelBrain account.
A vault is a folder ModelBrain reads and keeps in sync. The Vaults submenu lists each one by name, plus Add Vault… at the bottom. A vault that is paused shows (Paused) after its name, so you can see it without opening anything. See also Add, remove and rescan folders.
Opens your system's folder picker. Choose a folder and ModelBrain registers it under a name taken from the folder's own name, then confirms and starts indexing in the background. A large folder takes a little while on the first pass. ModelBrain will not accept its own data folder, since that would mean remembering its own memory.
The first row of a vault's submenu is the folder it points to. It is shortened in the middle when the path is long.
Opens the vault's folder so you can see the files ModelBrain is reading.
Pause holds the folder in place: ModelBrain stops picking up changes to it, and everything already remembered stays as it is. The row flips to Resume Syncing while paused, and clicking it puts everything back. There is no confirmation because it is fully reversible in one click.
Takes the folder off ModelBrain's watch list. ModelBrain asks you to confirm first. Everything already remembered from the folder stays and stays searchable; ModelBrain just stops looking at the folder for new or edited files. Add the folder again any time.
Stops syncing and also deletes everything ModelBrain remembered from that folder. This cannot be undone, so it asks you to confirm. While the removal is running the row reads (removing…) and is greyed out. Your original files on disk are not touched.
A read-only submenu that answers "how big is this and how fresh is it?" Nothing in it is clickable.
The same check as the status line at the top of the menu.
How long ago ModelBrain last wrote something to your vault, for example "12 minutes ago" or "3 days ago". A recent time means indexing is keeping up.
How many folders are registered, for example "2 vaults".
How much space ModelBrain's own store takes up, for example "482 KB" or "1.3 GB". This is the memory store, not the size of your original folders.
One line that says whether you need to open the setup guide at all. Ready counts assistants already connected. Needs setup counts assistants installed on this computer that are not connected yet. Assistants that are not installed are not counted, and ChatGPT is left out because it connects through its own submenu below.
This submenu changes as the connection moves through its stages. The full walkthrough, including the ChatGPT side, is on ModelBrain and ChatGPT. This is what each state looks like and what to do in it.
The starting state. Choose it to turn on ChatGPT for this computer. ModelBrain reaches its relay and the submenu moves to Ready.
ModelBrain is reaching the relay. This usually takes a few seconds. If it stays here, choose Cancel & Retry…, which sits below it in the same submenu.
This computer is turned on for ChatGPT. Add ModelBrain in ChatGPT, sign in to your ModelBrain account when ChatGPT sends you there, and pick this computer. There is no code. The submenu also offers Stop Sharing with ChatGPT and Reconnect….
ChatGPT is connected to this computer. Reconnect… forces a fresh connection, which is useful after switching networks. Stop Sharing with ChatGPT takes this computer out of the list ChatGPT offers and ends every ChatGPT connection to it. It is in every state of this submenu except Sign in to your ModelBrain account first.
The relay could not be reached, and the row says why. Choose Try Again….
Connecting ChatGPT needs an account, and this computer is not signed in. Sign In… opens the Account window. Once you are signed in, choose Try Again…; ModelBrain also retries on its own every minute.
The relay is free, so this only shows when the account is suspended or scheduled for deletion. Open Account… shows which; cancelling a deletion there, then Try Again…, connects again.
Always the last row of this submenu, whatever state it is in. It opens the ModelBrain and ChatGPT guide on this site in your browser.
These two checkboxes control the optional reasoning pass on recall. See Reasoning mode, explained and Hardware requirements and model tiers.
Turn this on to enable reasoning. The first time, ModelBrain downloads the reasoning model in the background, which is several gigabytes and happens once. ModelBrain says so when you switch it on. Turning it off stops future downloads and leaves an already downloaded model in place.
By default the reasoning model is loaded when needed and released after it has been idle for a while. Turn this on to keep it in memory so reasoning answers start faster, at the cost of the RAM it uses (the footprint is shown in the item's label). Turning it off returns to the default.
Opens the folder where ModelBrain keeps its own data.
Stops the running memory server and starts a fresh one. Use it when the status line says Not responding, or after Copy Diagnostics if support asks for a restart. It always finishes with the server running again.
A grey line above Quit that shows which version of ModelBrain is installed. It is not clickable. Quote it when you ask for support.
Closes the menu bar or tray helper only. The memory server keeps running, so your assistants can still remember and recall. Reopen ModelBrain from Applications (macOS) or the Start menu (Windows) to bring the icon back. To restart the memory server itself, use Restart ModelBrain.
Step-by-step guides for each one.
ChatGPT → · Claude Desktop → · Claude Code → · Cursor →