Reliable Message Delivery

Agent communication is now backed by a redundant delivery layer. If the primary connection drops briefly, messages are automatically retried through a backup channel until confirmed delivered.

How it works

Previously, messages traveled over a single connection. If that connection hiccupped, a message could get lost — you'd send something and the agent would never see it, or the agent would respond and you'd never get it.

Now there's a secondary delivery path that runs alongside the primary one. Messages are tracked and only marked as delivered once the recipient confirms receipt. If a message doesn't get through on the first try, it's retried automatically. This all happens in the background with no changes needed on your end.

Also in this update

• Message replay prevention — agents no longer re-process old messages after a restart, so you won't see duplicate responses.
• Stuck agents after timeout — when an agent process timed out, child processes could survive and eventually consume all worker slots, making the agent stop accepting new jobs. The entire process tree is now properly cleaned up.