mission control for
everything you run.

one place to talk to your agents, connect the accounts they need, and see what needs you.

get started

one agent above the projects

say it once. it lands in the right place.

maestro knows what every project owns, so a request goes to the project that already has the context and the accounts — or becomes a new project with the right agents from the start. you approve the plan, not the plumbing.

projects own the work

nothing is configured. everything is derived.

a project holds its conversations, context, credentials and history. its menu comes from what it can actually reach: connect a bank and accounts appear; disconnect it and they leave. what a project can do is evidence, never a setting.

you decide what ships

agents prepare. you approve.

anything that publishes, spends or trades waits for you, or stays inside limits you set in advance. every approval is recorded against exactly what was reviewed, so an edit after the fact needs a new decision.

it keeps going

quiet while it runs. loud when it matters.

scheduled agents, a watcher that wakes one only when a price crosses a band, and background passes that keep every project current. none of it asks for attention unless something needs you — and then it's on home.

runs where you are

agents run on your own machine, next to your files and your logged-in accounts.

your ai accounts

bring the claude, openai or google account you already have. usage is yours to see.

one identity per project

an agent acts as one project. reaching into another is a visible handoff, never a borrowed key.

connects to

plaid

what should maestro do?

get started