#52 Managed llama-server daemon keeps running and holding RAM after the CLI exits

[Anonymous]

Originally created by: sosidudku1

The managed llama-server is spawned as a detached daemon, so after closing the terminal it keeps holding RAM and VRAM until atomic-agent models stop is run manually. Staying warm between sessions is a fair default (model load takes tens of seconds), but there is currently no way to opt out.

Proposal, either of:

• a config flag like localModels.managed.stopOnExit that stops the chat daemon when the last CLI session exits
• an idle timeout that stops the daemon after N minutes without requests

Either would also help users on 16 GB machines where the idle model blocks other work.

Related: PR [#50] already adds daemon teardown when switching to an external server, so the stop path exists; this issue is about the exit and idle cases.

[Anonymous]

Ticket changed by: Ooooze

• status: open --> closed