| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| appcast.xml | 2026-11-17 | 28.1 kB | |
| Llama.dmg | 2026-11-17 | 1.3 MB | |
| 0.48.0_ Several models at once source code.tar.gz | 2026-11-17 | 314.7 kB | |
| 0.48.0_ Several models at once source code.zip | 2026-11-17 | 419.9 kB | |
| README.md | 2026-11-17 | 722 Bytes | |
| Totals: 5 Items | 2.1 MB | 0 | |
Models now stay loaded side by side for as long as they fit in memory. When a new one doesn't fit, the app unloads the one you used least recently to make room. This release also updates the engine to b12375, which is built on llama.cpp 1.0.
- Keep several models loaded at once, as many as fit in memory
- Update the engine to b12375, built on llama.cpp 1.0
- Show catalog names in the menu, such as Gemma 4 and GPT-OSS
- Show image and audio input markers on installed models
- Show the audio input marker on recommended models too
- Support video input when ffmpeg is installed with Homebrew
- Keep plus signs in quick prompts, so C++ no longer arrives as C
- Say that agents like Pi and OpenClaw don't need agent mode