0.21.0

• Update header to show base URL as a separate element
• Include memory budget details in context length info in Settings
• Add Hugging Face link to model context menu
• Remove --no-mmap to enable memory mapping
• Update llama.cpp to b7652
• Fix Nemotron KV cache footprint calculation

0.20.0

• Refine catalog menu layout and navigation
• Show granular download progress with decimal percentages
• Optimize settings menu toggle to avoid full menu rebuild

0.19.0

• Family items now open detailed views instead of expanding in place
• Add descriptions to model families
• Show incompatible models in the catalog with clear memory requirements
• Group Qwen3, Qwen3 VL, and Ministral 3 models with their reasoning variants

0.18.0

• Add Q4_K_M quantizations for Devstral 2 models
• Refactor catalog and model structures for better organization
• Update llama.cpp to b7569

0.17.0

• Add Devstral 2 model family
• Add Nemotron Nano 3 model family
• Use xcassets for icon to reduce bundle size
• Update llama.cpp to b7561

0.16.0

• Add Show in Finder button to installed model context menu
• Always show context length and estimated memory usage in model items
• Remove memory limit option in favor of automatic safety budget
• Add expandable info descriptions to settings
• Fix UI flicker when downloading or cancelling models
• Update llama.cpp to b7475

0.15.0

• Added "Memory cap" setting to replace hard-coded logic and allow user control
• Removed confusing right-click button for "run at max ctx"
• Renamed "Context length" option to "Context length cap" and updated labels
• Updated context length display in model metadata
• Made model deletion immediate for better UX

0.14.0

• Removed show quantized toggle — catalog now prefers full-precision models and falls back to quantized only when full-precision won't run on device
• Added setting to set context length
• Added setting to display estimated memory usage
• Redesigned model items and settings
• Updated icon for better integration with macOS Tahoe
• Updated llama.cpp to b7406

0.13.0

• Added Ministral 3 model family
• Added vision support indicator to catalog
• Replaced hover-based delete with right-click delete for installed models
• Added toggle button for quantized models in catalog divider
• Changed default context window from max to 4k
• Improved visual hierarchy with refined colors and font sizes throughout
• Updated app icon to better match menu bar icon
• Updated llama.cpp to b7247

0.12.0

• Enable resuming model downloads after network interruptions
• Add default sampling parameters based on model author recommendations
• Add experimental option to expose llama-server to network by binding to 0.0.0.0 -- see issue #17