Privacy First AI.

AI teammates know the room they're in — its messages, its artifacts, the decisions that happened there. They don't automatically know other rooms, even in the same workspace. They never know other workspaces.

If you want a teammate that spans rooms — for example, a research assistant working across multiple project rooms — you grant that broader access explicitly. Broader access is visible, not implicit. You'll see what each teammate can see, on the teammate's settings.

There's no hidden “AI memory” silently collecting context outside the rooms it's in. If we're remembering something, you can find where it came from.

AI proposes; you approve, edit, or reject. The approval card is visible in the room — the people doing the work see it, the decision is recorded, the AI cannot route around it.

Approval requirements are configurable per teammate, per room, per category of action. Low-stakes actions can run without approval if you choose. Consequential actions cannot.

We don't think of approvals as friction. We think of them as the contract that makes AI as a teammate possible.

Browser Control is bounded by three rules — together, they're what makes a browser-acting AI teammate something you can actually trust:

1. User-enabled. The extension is installed and turned on by you. We don't silently activate.
2. Room-scoped. Browser Control activates from a specific Sociail room — not from the browser tab.
3. Per-action approved. Every action gets its own approval card. Every time. There is no “approve once, run forever” mode.

Sociail does not read browser pages in the background. Sociail does not use the extension to monitor your browsing. The extension only acts when you've approved a specific action — and you can see exactly what it did, after.

Sociail uses third-party AI model providers — like Anthropic — to power AI teammates. We're transparent about this because anyone reading carefully would figure it out anyway.

When you direct a prompt to an AI teammate:

— The necessary inputs are sent to the provider for that request, and only that request.
— The provider processes the request and returns a response.
— The provider is contractually prohibited from using your content to train their general models.
— We don't send more than the request needs to fulfill itself.

We don't pretend providers see nothing. We do make sure providers can't keep what they see — for their own models, or anyone else's.