Workspace
Fleet — Coordinate Your AI Agents
Organize connected AI agents, assign responsibilities and keep agent-driven architecture work visible in one coordinated environment.
Keep connected agents and their work in one place.
Organize connected AI agents, assign responsibilities and keep agent-driven architecture work visible in one coordinated environment.
Fleet shows agents that are already connected, the teams they belong to, the roles that describe their work, and whether a deployment is waiting. It links decisions in the AI-native IDE with the execution record those agents report back.
Fleet does not start an external agent. Handing work to an agent through MCP is not the same as launching that agent. The current model is pull-based: an authorized agent retrieves the work that was prepared for it.
Connect and authorize an agent
Connect an agent with the supported MCP configuration. See Connect an AI Agent.
Authorization is per project. An agent can be connected to the account and still be unauthorized for the project you are viewing. Read, deploy, and interactive permissions stay separate. The deploy tool is granted only when that permission is present.
Agent Dock
Agent Dock lists each connection, the client it connected from, whether it is authorized for this project, the tools it may use, any fleet role, and whether it has an active deployment.
Refreshing the dock reads the current connections. It does not start an agent. A connected agent is not started from this page.
Teams, roles, and work packages
Assign an authorized agent to a collaboration team. The assignment records who is expected to work with that team. It does not grant a new security permission, and it does not create a deployment by itself.
Roles describe responsibility: a name, a short description, detailed instructions, responsibilities, and expected outputs. Saving a role does not give the agent access it does not already have.
Architecture changes that should be implemented are prepared as deployment work packages in the AI-native IDE. A connected agent with the project in its authorization, the MCP deploy permission, and the claim tool can retrieve a package and report progress. See Deploy on Agent.
Status, results, and tokens
Fleet and the deployment log distinguish work that is waiting, claimed or running, completed, failed, or canceled. Open the reported result to read what the agent submitted.
A successful agent report does not by itself prove that the infrastructure changed as intended. Use Update Discovery when you need to refresh project evidence.
One accepted deployment request consumes one deployment token. Preparing an assignment or a role does not consume a token.
How Fleet fits the IDE
Use the IDE to decide the architecture change and to open Deploy on Agent. Use Fleet to see which agents are authorized, which tools they hold, and which deployment is active. Fleet does not replace the deployment dialog, and the deployment dialog does not start a remote agent.
Availability
Fleet is included in Essentials, Growth, and Architect.
Important to know
Important. SPECTRALE does not wake or launch arbitrary external agents. An agent must already be connected and must pull the work it is allowed to see.
There is no universal autonomous execution. An agent without the required permission cannot perform that action.