Local AI Workflows

Where Did 120GB of Mac Storage Go?
Check Your Local AI Models.

Running local LLMs on Apple Silicon is incredible—until you realize three versions of Llama 3.1, a DeepSeek checkpoint, and two Flux diffusion models just consumed 100GB of your SSD. DiskWarren maps hidden SHA blobs back to human names so you know exactly what to keep.

Ollama Model Blobs (~/.ollama/models)

Ollama stores weights as SHA-256 content hashes (e.g. sha256-6a0746d16...) inside hidden subdirectories. When you open Finder, it looks like an unidentifiable 35GB blob. DiskWarren parses the manifest hierarchy to show the real model tag, context length, quantization, and disk footprint.

LM Studio GGUF Quantizations

It is common to download multiple quantizations of the same model to test speed (e.g. Q4_K_M vs Q6_K vs Q8_0). DiskWarren clusters variants of the same base architecture together, making it simple to keep the fastest one and recycle the redundant 15GB files.

Hugging Face Hub Symlink Caches

Hugging Face caches checkpoints in ~/.cache/huggingface/hub using Git LFS pointer symlinks that point into a shared blobs/ folder. Deleting a snapshot folder in Finder doesn't free the disk space—DiskWarren resolves the actual underlying blobs and cleans them cleanly.

ComfyUI Checkpoints & LoRAs

Image generation pipelines hoard massive weights: SDXL, Flux Schnell, Flux Dev checkpoints (12GB – 24GB each), plus dozens of downloaded LoRA adapters. DiskWarren indexes your ComfyUI models directory with last-used timestamps so you can clear dormant experiments.

Check Your AI Storage via Terminal

Curious how much space your local AI tools are holding right now? Paste these into Terminal:

Terminal AI Storage Audit
# Check Ollama storage size du -sh ~/.ollama/models # Check Hugging Face hub cache size du -sh ~/.cache/huggingface/hub # Check LM Studio downloaded models du -sh ~/.cache/lm-studio/models 2>/dev/null || du -sh ~/Library/Application\ Support/LM-Studio/models 2>/dev/null

Take Control of Your AI Storage

Scan your Mac for free. Correlate Ollama blobs and discover duplicate GGUFs in seconds.