Agent sharing

When you create an agent, you own it. You decide which tenants (and through them, which teammates) can put it to work on their projects.

How sharing works in the UI

Sharing lives on the agent, not on the device.

1. Go to Settings → Agents. Find the agent you own.
2. Click Share. A list of your tenants appears with a checkbox each.
3. Check a tenant to share the agent with its members. Uncheck to revoke.

That’s it. The status line under each agent shows its runtime and which tenants currently have access.

Removing a share doesn’t kill running tasks. The agent finishes what it’s already executing. To pull the agent off a specific project, go to Settings → Projects → Agents and detach it there.

What the share actually grants

Members of a checked tenant can attach the agent to their projects. Once attached, anyone with project access can assign tasks to that agent the same way they’d assign to a human teammate. The agent executes on the device its owner paired, so the agent’s owner still controls what hardware the work runs on.

In other words: sharing is about who can dispatch work. The physical execution always happens on the owner’s device.

What sharing is NOT

• Not a shared login. Teammates use their own accounts and their own permissions. Audit logs show them as the assigner, the agent identity as the executor.
• Not a shared session. Teammates don’t see your UI or your boards, only the projects they themselves are members of.
• Not workspace-wide by default. Each agent shares independently. You can share one agent broadly and keep another private without touching the second one.

A note on terms-of-service

When an agent’s runtime is bound to a personal subscription (e.g. a Pro/Max coding-agent account), the device is automatically marked owner-only and sharing is disabled. This is to stay on the right side of provider terms-of-service that forbid using a personal subscription as a multi-user backend.

For shared agents, use API-key / programmatic auth. The product enforces this so you don’t have to think about it.

See also

• Auto-pickup and dependencies: once an agent is shared, auto-pickup lets it claim ready work without anyone dragging cards.
• Headless teamwork: for shared compute, run a headless worker and share its agents across the team.
• Audit trail: every assignment, every execution, every permission decision is recorded with the identities involved.