MLX Agent
Discover, adopt, and safely wire local MLX models on Apple Silicon.
Product pageModels
5 public models under huggingface.co/cavi-ai (opens in new tab), listed from the Hub as of Nov 14, 2026. Each model card states its measured results, requirements and license.
| Model | Task | Base model | Format | Size | Downloads, 30 days | Download |
|---|---|---|---|---|---|---|
| Qwen3-VL-8B-Instruct-MLX-4bit (opens in new tab)Apache-2.0 weights | Vision-language | Qwen/Qwen3-VL-8B-Instruct (opens in new tab) | MLX 4-bit | 5.78 GB | 127 | hf download cavi-ai/Qwen3-VL-8B-Instruct-MLX-4bit |
| bobby-control-detector (opens in new tab)Apache-2.0 weights | Object detection | PekingU/rtdetr_v2_r18vd (opens in new tab) | safetensors + ONNX | 0.16 GB | 58 | hf download cavi-ai/bobby-control-detector |
| Audio8-ASR-Infinite-MLX-8bit (opens in new tab)Apache-2.0 weights · MIT code | Speech recognition | Edge0/Audio8-ASR-Infinite (opens in new tab) | MLX 8-bit | 5.29 GB | 48 | hf download cavi-ai/Audio8-ASR-Infinite-MLX-8bit |
| Audio8-ASR-Infinite-MLX-4bit (opens in new tab)Apache-2.0 weights · MIT code | Speech recognition | Edge0/Audio8-ASR-Infinite (opens in new tab) | MLX 4-bit | 3.75 GB | 38 | hf download cavi-ai/Audio8-ASR-Infinite-MLX-4bit |
| laya-MLX-8bit (opens in new tab)Apache-2.0 weights · MIT code | Text classification | convaiinnovations/laya (opens in new tab) | MLX 8-bit | 1.62 GB | 500 | hf download cavi-ai/laya-MLX-8bit |
Measured results on an Apple M5 Max for each MLX conversion are on the local models page.
Discover, adopt, and safely wire local MLX models on Apple Silicon.
Product pageNative macOS app for converting, verifying, serving and comparing local MLX models.
Product page