Loach v1.4.1

Fixes

• Fixed context menu backgrounds. Context menus are now easier to read.
• Reply stats stay put. The token count and speed under a finished reply no longer disappear the moment it stops streaming, only to reappear after a restart.
• A model suggestion you can actually download. When nothing in the catalog fits your free disk space, setup now points at the smallest option instead of headlining the largest one with its own Pull button disabled.
• No stale OpenAI models. Removing your API key now clears the model list, rather than leaving entries in the pickers that could only fail with an auth error at send time.
• In-chat find agrees with itself. Highlights now match the result counter exactly, so Next and Previous reach every one — expanding an imported conversation no longer leaves highlights the counter never counted.
• Focus finds its way back. Pressing Cmd/Ctrl + K while the search palette is already open no longer strands focus on the page when you close it; it returns to the composer.