Cua Cloud Fleets Adds Auto-Expiring TTLs for Agent Sandboxes
- Source
- x.com
- Date
1/ Cua Cloud Fleet resources can now clean themselves up. Set a creation-age TTL on a pool or claim, and Fleet deletes it automatically when the deadline arrives. Useful for CI, eval runs, and temporary agent workloads. Launch a Cloud Fleet today: https://t.co/1qqkbXXZ4o
2/ The same field works at both levels: Pool: ttl_seconds_after_created=86400 Claim: ttl_seconds_after_created=3600 No TTL means no automatic expiry. The pool or claim stays live until you explicitly delete it.
3/ The clock starts at creation and not at last use. Reapplying a named pool does not reset its age. Reconnecting to a named claim keeps its original deadline.
4/ One important boundary: if a pool expires while a claim is still held, Fleet drains the pool. That claim's sandbox is destroyed when released instead of returning to warm capacity. Docs: https://t.co/mQSMecAHE1
Context
Cua says its Cloud Fleet pools and claims can now carry a time-to-live. In Cua's thread, the setting is a creation-age field, ttl_seconds_after_created, that works on both a pool and a claim, and Fleet deletes the resource automatically when the deadline arrives. Cua names CI, runs and temporary workloads as uses. With no time-to-live set, there is no automatic expiry and the resource stays live until someone deletes it.
The clock starts at creation, not at last use, so reapplying a named pool does not reset its age and reconnecting to a named claim keeps its original deadline. Cua also flags one boundary: if a pool expires while a claim is still held, Fleet drains the pool, and that claim's sandbox is destroyed when released instead of returning to warm capacity.
- sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.
- eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Checking sign-in…
Loading comments…