Every agent run in its own pod
Each agent run now starts in its own isolated pod, apart from Veyra's API and from other runs, and the pod is removed when the run ends.
Agent runs on Veyra Intelligence no longer share a machine with Veyra itself. Every run gets a place of its own.
- One pod per run. When an agent starts a run, Veyra starts a new, isolated pod for that run alone, with the agent's tools and its workspace. Runs from other agents and other organizations never share it.
- Nothing runs on the API host. The commands, tools and code an agent runs stay inside its pod and never run next to Veyra's API.
- Removed after the run. When the run ends, its pod is deleted. Pods left behind by an interrupted run are cleaned up too.
- Busy times wait, not fail. When many runs start at once, a run waits for room to start its pod instead of failing.
Nothing changes in how you create agents, assign tasks or read a run's output.