AI coding assistants have become a standard part of many developers' workflows. Windsurf, Codeium, and similar tools run inference in the background, index your code, and maintain local state — all of which creates storage that accumulates over time.
What these tools store
AI-powered code completion and suggestion tools typically maintain some combination of:
- Local inference caches — results of recent AI completions, kept to speed up repeated patterns
- Code indexes — structured representations of your project for faster, more relevant suggestions
- Workspace state — per-project configuration and context
- Model files — if the tool runs models locally rather than calling a remote API
How much storage this represents depends heavily on how many projects you work on, how large those projects are, and whether the tool uses local models.
Want to skip the hidden-folder hunt?
LLM Cleaner scans your Mac for local AI models, caches, indexes, and project memory — then shows what you can review, reveal, export, or safely move to Trash.
The local model question
Some AI coding tools use cloud APIs by default but can be configured to run models locally. If you have enabled local model inference at any point, those model files are probably still on your machine — and they are typically among the largest files an AI tool creates.
Unlike inference caches and indexes, local model files do not rebuild automatically if deleted. Removing them means re-downloading them if you want to use local inference again.
The right question to ask first
Before deleting any cache from an AI coding tool, the useful question is: is this from current active projects, or from projects I finished or abandoned?
Old project caches and indexes are the safest to clean. Current project storage — especially AI context files — is worth reviewing before removal.
Not sure what is safe to delete?
LLM Cleaner separates models, rebuildable caches, and project memory so you do not treat everything like junk.
Why deleting blind is a bad idea
It is tempting to find the largest folders in an AI tool's storage and delete them to reclaim space. The problem is that storage folder structures are designed for the tool, not for human readability. The folder that looks largest might be a shared component cache that every project depends on, not an abandoned old project's index.
Deleting it silently degrades your coding assistant until the cache rebuilds — which may take time you did not plan for in the middle of active work.
Local AI storage is messy. Cleanup should not be.
Find Ollama, LM Studio, Hugging Face, Cursor, Windsurf, Claude Code, Whisper, and loose model files in one place.