The agent's scratch disk and its traffic are already paid for

A research agent is messy. It downloads a 6GB dataset, unpacks it to 40GB, writes intermediate parquet files, pulls a dozen models to compare, and serves a 300MB result to the user who asked.

On platforms that bill storage per GB-month and egress per GB, a thorough agent is an expensive agent. You end up writing cleanup logic whose only purpose is keeping the agent's curiosity off the invoice.

On Hoody the container's disk is NVMe on the server you rent, and bandwidth and storage are included in the server's monthly price. No tiers, no per-GB line, no egress charge when the user downloads the result. Need more room than the tier offers? The fix is a larger server tier, not a storage product.

Disk and bandwidth came with the box; the agent can't run up a bill by being thorough.