4.0.0

4.0.0

New

• llama.cpp provider — Horme now talks directly to llama-server, including router-mode setups with lazy model loading. Configure a chat URL/model and a separate embedding URL/model in Settings; llama.cpp is available everywhere other local providers are: chat, tag generation, tag translation, Vault Brain (RAG) embeddings, and local fallback.

Fixed

• Model and preset dropdown buttons no longer overflow their box when the panel is narrowed — long names now truncate with an ellipsis.
• The connection status indicator no longer gets stuck red after the provider comes back online — it now re-checks automatically every 15 seconds instead of only on view open or manual refresh.