StudioAgents work through Studio:Blueprint, not around it.
Every agent action is scoped, logged, and consultant-approved. Here is exactly how.
The four permission modes
Every agent runs in one of four modes. The mode is set by the consultant and visible in the product.
The agent reads firm and client data. It proposes nothing and writes nothing.
Surface a risk from the Alert Stack.
Create a roadmap action.
The agent suggests actions. Nothing is committed until the consultant approves.
Draft a finding for the Evidence Canvas.
Attach it without approval.
The agent can write to specified surfaces after explicit per-action approval.
Log an approved decision to the Decision Ledger.
Publish a diagnostic without consultant sign-off.
The agent can act autonomously within a narrow, defined scope the consultant sets.
Send a scheduled Progress Room snapshot to a client.
Change a score or add an engagement.
The propose-only default
Every new agent starts in propose-only mode. The consultant sets the permission level for each agent explicitly. There is no global autopilot switch.
Every proposal is logged
Every agent proposal is logged. Accepted proposals carry the approver's ID, timestamp, and the methodology version active at approval.
Rejected proposals are retained in the audit trail.
Bring your own agent
Studio:Blueprint exposes its data and actions via MCP. Bring your own agent. Claude, ChatGPT, OpenClaw, or any MCP-compatible agent.
The permission model applies regardless of which agent is connected.
Want to use agents with Studio:Blueprint? Download the Agent Pack →
See the rest of the model
The permission model sits inside a wider governance framework. The demo shows how it runs in practice.