Teammates, roles & invites

A workspace is a team. This page covers adding people and setting their level of access.

Roles#

• Members (daily users) chat with agents, run missions, review deliverables, and approve actions. They cannot reach operator surfaces.
• Operators (owner / admin) do everything members can, plus create agents, set policy, manage skills, integrations, secrets, and the governance surface.

Note: The role boundary is enforced on the server, not just hidden in the UI. Admin, config, and destructive actions are gated server-side, a member can't reach them by guessing a URL. Operators get a "preview as daily user" toggle to see exactly what members see.

Inviting people#

Create an invite link from the members menu in the sidebar. One link, a few choices: a role, an expiry (7 days by default), an optional email to bind it to one person, and what it sets up on their machine, AI-tool discovery, and whether their coding agent is governed too (see Shadow-AI discovery). Whatever you choose, the app discloses it plainly on their machine before anything starts.

• Email-bound (recommended). Bind the link to a person's email. Only someone who signs in as that exact email can join. A leaked link is useless to anyone else.
• Open. Leave the email blank for a link anyone can use to join until it expires. The dialog flags this clearly, and it is still governed by your registration mode below.

Ceiling: Access is bound to a verified sign-in, never to a typed-in address. An email-bound invite is accepted only when the authenticated identity's email matches the link, so a self-typed email can never, by itself, grant access. Invite links are strong random tokens, rate-limited, and scoped to a single workspace. Setting up discovery only inventories a machine's AI tools, it never grants workspace access.

Who can join#

Set the workspace's registration mode in Settings → Workspace:

• Open: anyone can sign up.
• Invite: only people with a valid invite link (the default).
• Closed: no new signups.

Authentication is email + password (12-character minimum) or Google sign-in.

More than one workspace#

A workspace is a separate room, its own agents, channels, rules, and audit trail, isolated from every other. Teams use a second one to separate a department, a client, or an experiment from production.

Creating one is operator-only by default. You can let members create their own from Security → Team, and there's a reason it isn't on out of the box: without a rule, "make a new workspace" is the shortest route around whatever you decided in the old one.

So a new workspace inherits the policy of the one it was created from. Not a copy you have to remember to apply, governance is on from the moment it exists, at the posture you already set. A separate surface, never separate governance. The creation itself is recorded in the parent's audit trail.

Ceiling: Isolation cuts both ways, and this is the honest limit: a parent workspace cannot see into a child. You'll know one was created and that it runs under your policy, you will not see its Registry, its Trace, or its channels from outside it. If you need consolidated oversight across many workspaces, talk to us rather than assuming it's there.

Next#

• Agents & scheduling, creating the AI side of the team.
• Policy & enforcement, what members are allowed to do.