Direct Repository Access: Search for models on Hugging Face directly within the app.
Repo ID Support: Paste any Hugging Face GGUF repo ID to fetch available files instantly.
System Diagnostics
VRAM & GPU Check: Auto-detection of available NVIDIA/AMD VRAM.
Compatibility Badge: Flags models as "GPU Compatible" or warns about "Insufficient VRAM" before you download.
Background Download Manager
Parallel Downloading: Several GGUF downloads can now run at once (previously a second download was rejected outright), with combined progress in the engine status bar and a rough time estimate, not just a bare percentage.
Lifecycle Control: One-click Start, Stop, and Update for all local models.
Hardware-Matched First-Run Suggestion: Setup reads your RAM/GPU and recommends a matching chat + memory model pair, with one button to start both downloading.
Safer Context Sizing: The context-size field now reads the model's real maximum context straight from the GGUF file and won't let the slider exceed it (previously free-text, could crash the engine).
Accurate Capability Badges: Tool-calling/code badges are now derived from the model's actual chat template and tags instead of a hardcoded list of "known" families.
Plain-Language Errors & Tooltips: Raw errors like llama: server failed to become ready within 120s are now short and actionable; hardware-fit and quantization badges have hover tooltips explaining what they mean.
Discover Filters: Tools/Vision/Code/Embedding/Size filters now combine with OR (not AND) and are grouped into multi-select dropdowns with an "N filters active · clear" indicator.
