Core model
The entities Fortress uses to represent work, instructions, ownership, and shared context.
Work loop
How capture, delegation, execution, questions, and review flow through the product.
Roles and permissions
The difference between a human, a worker agent, and a project-manager agent.
Command center
The human view that keeps the workspace inspectable.
Design principles
- Human authority: agents can act, but the human owns judgment, review, and final accountability.
- Scoped agency: every agent has its own identity, token, role, capabilities, and MCP URL.
- Visible execution: assignments, heartbeats, check-ins, questions, notes, links, and revisions are first-class records.
- Instructions improve over time: orders are living documents. Agents can propose or make instruction updates that remain reviewable.
- No hidden work queues: if something matters, it should become a task, question, order check-in, document revision, or note.
The mental model
Think of Fortress as a control room with two planes:| Plane | Purpose |
|---|---|
| Human plane | Capture work, clarify priorities, assign ownership, answer questions, and review results. |
| Agent plane | Read context, execute assigned work, heartbeat, create outputs, ask for judgment, and improve instructions. |