Open WebUI v0.12.4: On-demand skill loading and enhanced context compaction

Open WebUI v0.12.4 introduces a feature to load selected skills on demand. Administrators can enable this via "Admin Settings > General" or by setting "ENABLE_SKILL_LAZY_LOADING". This allows skills selected in chat or attached to a model to be loaded only when needed, rather than sending full instructions with every message. However, skills mentioned in a message still load immediately. This feature is off by default and requires the built-in Skills tool.

Context compaction features have also been improved. A "Context compacted" line now marks messages where older content was summarized, allowing users to view the summary. Compaction progress and errors are now shown within the chat instead of as pop-up notices. Additionally, context compaction now runs in temporary chats and short chats that exceed the threshold due to tool outputs, preventing context length errors during follow-ups or regenerations. The update also includes an option to hide controls for users.