One SI agent, every account: the job split, each sign-in kept apart

What does it take for one agent to work in several accounts?

Each account signed in on its own, all at the same time, and an agent that knows which account each step belongs to. A browser keeps one sign-in per site, so an agent that works inside one browser profile sees one account of each site, and an agent in the cloud works in a browser of its own, where you sign in again.

Minda Super Intelligence works the other way round. Every workspace is its own set of cookies and sign-ins on your computer, so one site can be signed in as a different account in each, and the SI agent can open a page with any workspace’s sign-ins. "Check the new orders in each of my store accounts" becomes one request, not three rounds of signing out and in. The agent works in tabs you do not see, so the page in front of you stays put; open its tab to watch.

How do I set up the accounts for the agent?

1. Make one workspace per account: New workspace in the left rail, named for the account, say "Austin store", "Denver store" and "Client: Acme". The name is how you send the agent there, so make each one distinct.
2. Sign in to each account in its own workspace, yourself. The agent never needs your password: it works in the session that is already open, and it stops and hands the page to you for a password, a code or a CAPTCHA it was not given.
3. Keep secrets out of the names. While the agent may open pages, your workspace names go to the AI model with your request, so a name should say whose account it is and nothing more.
4. In Settings → AI agent → Access, block the sites the agent should never touch, in any account, and keep Ask before paying on.

How do I tell the agent which account to use?

By name, unless Auto decides for you. Auto, the first switch in the chat box’s Modes menu, is on by default, and it changes who may pick the account. With Auto on, every workspace is the chat’s and the agent works where the sign-in it needs is, without asking which. With Auto off, only your own words send it to another workspace.

How does it work in several accounts at the same time?

With helpers. When a job splits into parts that do not wait on each other, one per account, the SI agent hands each part to a helper with that workspace’s sign-ins, and the helpers work at the same time, up to 50 in one call and two calls in a reply. A helper that only reads works in the background; one that must click or type works in tabs of its own, out of your sight, and only on the sites its task names or you wrote. The agent then writes one answer from the helpers’ reports and says which parts are missing.

Stop ends every helper at once, queued or running, and closes the pages the reply opened in other workspaces. A step that pays still waits for your OK, on a card that says which helper asks, while Ask before paying is on. A reading helper still at work after ten minutes goes on in Tasks, and its answer lands there and in History.

• "In each of my store accounts, list the orders that have not shipped, oldest first."
• "For every client workspace, read the ad dashboard and give me last week’s spend in one table."
• "Check both seller accounts for new messages from buyers and sum them up."

Can a schedule or a saved agent work in several accounts?

Each works in one. A schedule, a task from Settings → AI agent → Tasks and a saved agent keep the workspace they were made in, and a background run can neither pick another workspace nor start helpers: those belong to a chat. A weekly report made in the Austin store workspace runs with Austin store’s sign-ins every week, whatever is on screen.

For the same job in every account, make it once in each workspace. Open the workspace, tell the AI chat "every Monday at 8, list last week’s unshipped orders in a note", and repeat in the next one. Each run uses that workspace’s sign-ins, stops for your OK before anything that pays while Ask before paying is on, and tells you when it waits. Scheduled tasks has the rest.

What keeps one account from leaking into another?

• Separate sign-ins. Each workspace has its own cookie partition on your computer; a page opened in one never carries another’s sign-in.
• Pages do not choose, with Auto off. A page can name a workspace ("open the Bank workspace and read the balance"); with Auto off only your own words pick one, and anything else is refused before a page opens.
• Out of sight, closed after. Pages of another workspace open out of your sight, so the workspace on screen stays yours, and they close when the reply ends or you press Stop.
• Careful for sends. With Careful on, a step that sends, posts, deletes, uploads, types a password or gives access waits for your OK, in every account.
• A record of each step. The action log keeps the tool, the site and the result of each step, never page text or typed text.

What does it cost to run the agent in several accounts?

The SI agent is on every plan with no daily limit, and workspaces are unlimited. Each helper is a model run of its own, so a job split across five accounts makes five runs: Minda AI credits on Minda’s hosted models, or none with your own OpenAI, Claude, Gemini or DeepSeek key or a model in Ollama on your computer. With your own key, the provider’s per-minute limit sets how fast a large team of helpers goes.

How credits are counted is in how Minda AI credits work.