Run the work
Agents can build software, execute scripts, query data and run experiments. They inspect the output and decide what to try next.
Prepare once, use across the fleet
Start from a prepared Linux image. Add your dependencies and tools, then reuse that environment across agents and future runs.
CPU sized for the work
Each agent has its own computer. CPU capacity is shared, with allocations sized for the tools it needs. Tool execution alternates with model responses, so many tasks can use fractional CPU allocations. Larger builds and analyses can use larger environments.
Room for different workloads
Combine smaller environments with larger machines for memory-intensive analysis or builds. The fleet can share work between them.