opbox

People & Teams Governance

Run the people, AI agents, and teams in a workspace - who is on the roster, who reports to whom, and what each is allowed to touch.

A workspace is run by people and by AI agents, side by side, and both need governing. This feature is where you manage that roster: invite a colleague, set what tier they hold, build the agent org chart that the AI workforce reports up through, gather people into teams, and grant or revoke fine-grained access to specific things. Every one of those moves is a named, permission-checked, audited action, so the question “who can do what here, and why” always has a recorded answer rather than a guess.

What it does

A human roster with a clear lifecycle. Invite a person and they join the roster at a starting tier. Their tier moves between MEMBER, ADMIN, and OWNER as their responsibilities change, and the system blocks the one move that would lock everyone out: you cannot demote the last OWNER away. When someone leaves, deactivating them disables the account and bumps their session version so existing sessions stop working at once; reactivating brings them back, but does not silently restore the old sessions, so a fresh login is required. Each person can set their own avatar, surfaced back through their session.

An agent org chart with a chief-of-staff at the top. The AI workforce is not a flat pool of bots, it is a real org chart. A chief-of-staff sits at the top, with HEAD and WORKER nodes reporting up through it. Adding an agent node mints the user, gateway, key, and SOUL behind it in one reconciled motion, so the node and the running agent stay in step. You can re-parent a node under a different head, promote it to a different role tier, disable it, or remove it. The chart enforces its own shape: no cycles or orphans, never a second chief-of-staff, never a demotion of the sole chief-of-staff away from that role, and the chief-of-staff node cannot be removed or self-removed.

Power bounded by structure, not by trust. An agent’s reach is set by where it sits in the chart and by the guards on each move, not by how much it is trusted. Re-parenting and creation are lattice-guarded (the parent must be a chief-of-staff or a HEAD, and moving a HEAD needs chief-of-staff or admin authority), and the sensitive moves (create, move, remove) sit at the OWNER permission tier. The org chart is the shape that decides who answers to whom; the per-actor access and autonomy controls bound what each one can actually touch.

Teams to group the people doing the work. Gather people into a team, add and remove members, rename it, and archive it when it is done. A team is the lightweight grouping you use to organise who is working on what, alongside the formal roster tiers.

Fine-grained access grants. Beyond the tier someone holds, you can grant access to a specific thing and revoke it again. An access grant is recorded like everything else, and you can list the grants that are in place, so a one-off “let this person see this matter” is an explicit, reversible, auditable act rather than a quiet exception.

Break-glass controls for an incident. When something has gone wrong, two reversible emergency controls cut the org back fast. Freezing the org mass-disables the agents and bumps every session at once, stopping the AI workforce and forcing everyone to re-authenticate. Lockdown goes further: it does everything freeze does and also denies external egress and revokes the fan of token scopes, the maximal reversible cut short of taking the whole box down. Both are reversible, both are recorded with a parent record and a child record per affected agent and session, and both sit at the OWNER tier.

How you use it

Bring someone onto the workspace. Invite the person and they land on the roster at a starting tier, with a forced password change on first login. As their role grows, set their tier up to ADMIN or OWNER; when they move on, deactivate them and their sessions stop working immediately.

Build and adjust the agent org chart. Add a HEAD or WORKER node and the reconcile mints the user, gateway, key, and SOUL behind it. Re-parent a node under a different head, promote it to a new role tier, or disable a node you want to take out of play. Read the chart back and look up any single node to see where it sits and what it is.

Organise people into teams. Create a team, add the people who belong on it, rename it as the work shifts, and archive it when the work is done.

Grant and pull back specific access. Grant a person access to a particular thing, list the grants in force to see what reach has been handed out, and revoke a grant when it is no longer needed.

Cut the org back in an incident. If you need to stop the AI workforce in a hurry, freeze the org to mass-disable the agents and invalidate every session at once. If you need the maximal cut, lock the org down to also seal external egress and revoke the token scopes. Both are reversible once the incident is handled, and both leave a full record of what was cut.

The kernel verbs behind it

Each goes through the one front door - permission-checked and audited before it runs (see Security & permissions). These are the verbs most specific to people and team governance.

The human roster:

  • actor.invite - invite a person onto the roster at a starting tier.
  • actor.setRole - move a person’s tier between MEMBER, ADMIN, and OWNER (blocks demoting the last OWNER).
  • actor.deactivate - offboard a person: disable the account and invalidate their sessions.
  • actor.reactivate - re-enable a deactivated person (a fresh login is still required).
  • actor.get / actor.list - look one person up, or read the whole roster.

The agent org chart:

Teams and access:

Break-glass:

  • org.freeze - mass-disable the agents and bump every session at once (reversible).
  • org.lockdown - org.freeze plus a deny-all on external egress and a revoke of the token scope-fan (the maximal reversible cut short of taking the box down).

See the full set in the org verb reference, the actor verb reference, and the team verb reference.