LocalAI packages reorganized

Posted: 2026-10-04 by Plamen K. Kosseff | Revision: 1
Format:
2.0
Display if installed:
sci-ml/local-ai
Display if installed:
sci-ml/localai
The LocalAI packages of this overlay have been renamed and recategorized. Package moves migrate installed systems and world entries automatically:

sci-ml/local-ai            -> sci-ml/localai
app-local-ai/backends-meta -> www-apps/localai-server
app-local-ai/<backend>     -> localai-backend/<backend>
The two top-level names also swap roles. sci-ml/localai - the atom your world file now holds - is a meta package: it installs the server (www-apps/localai-server) plus the inference backends selected by its USE flags, which all default to on. The server package itself no longer references any backend package. Removing sci-ml/localai from the world file lets emerge --depclean remove the server and every backend in one sweep.

If you run a full LocalAI stack, no action is needed: the next world update follows the moves and keeps everything installed.

If you run the server WITHOUT backends, pick one before your next world update:

  • keep sci-ml/localai in world and disable its backend flags, e.g. echo "sci-ml/localai -*" >> /etc/portage/package.use/localai


  • or anchor the server directly: emerge --deselect sci-ml/localai emerge --noreplace www-apps/localai-server


Otherwise the meta package's default USE flags will schedule every backend for installation.

If the update aborts with file collisions on the server's files (/usr/bin/local-ai and the service files), unmerge the old server install - it carries the meta's new name in the package database - while keeping the world entry, then run the update again:

emerge -C --deselect=n sci-ml/localai
emerge -uDN @world
This is safe: the update reinstalls the server and the backends immediately, and /var/lib/local-ai (models, state) is untouched.