Ten thousand headless runs a day, no meter on any of them

Every support ticket gets an agent pass: summarize, classify, draft a reply, flag the ones that need a human. Ten thousand tickets on a busy day.

Your handler POSTs to /api/v1/agent/headless/runs with the prompt and gets a job_id back. The run happens inside the container's agent service; when it finishes, GET /jobs/{job_id}/result returns the output. No session to manage, no worker to keep alive.

On a per-invocation platform, ten thousand runs is ten thousand billable events plus their duration. On Hoody the agent service runs in a container on a flat-rate server; the ten-thousandth run is handled by the same process as the first, on cores you already rent. Tokens are separate: your own key invoiced by the provider, or the Hoody gateway at 5% over provider cost.

Invocation count stopped being a number anyone watches.