Fleet & nodes

A node is a dedicated Linux machine assigned to your company, running a persistent CCD agent. Your code lives on your nodes, work executes on your nodes, and nobody else's workloads share them. Your plan determines how many you get: 5 on Team, 12 on Scale, full-fleet on Enterprise.

Fleet overview

The fleet views show every node with its live status:

• Online: healthy and ready for tasks
• Offline: unreachable; tasks route to other nodes
• Health details (load, capabilities) are visible per node

You normally don't have to think about which node runs a task. CCD picks a healthy one. When you care, you can target a specific node.

Ways to run work across the fleet

• Dispatch: send a task to one chosen node
• Broadcast: send the same task to many nodes at once
• Pipeline chaining: chain agents so one node's output feeds the next (the Agent Chain page)
• MULTI mode: fan a single agent task out to every online node with one session tab per node (see The Agent)

GPU nodes

Nodes with GPUs are detected automatically (GPU type, VRAM, CUDA capability) and unlock GPU task types like ML training and rendering. GPU-aware features (for example the simulations in the Arcade) automatically adjust quality to the hardware they run on.

Folder locks

When a task is locked to a specific project folder, CCD serializes access: multi-node fan-out collapses to a single node so two agents never write to the same project at the same time.