One tool can behave like several applications
An AI coding environment is more than a text editor. It may run language servers, extensions, terminals, local databases, agent sessions, indexes, browser previews, package managers, and build tools. Each subsystem creates its own storage, often in a different part of your home Library or project directory.
The result is distributed growth: no single folder looks alarming, but the combined footprint of application data, caches, dependencies, and generated output becomes significant.
The main sources of storage growth
- Editor caches, logs, crash reports, workspace state, and extension storage
- Project dependencies such as node_modules and package-manager caches
- Build products such as .next, dist, build, coverage, and DerivedData
- Local databases, indexes, previews, downloads, and temporary artifacts
- Multiple experiments that duplicate the same dependency trees and generated files
Why Finder alone can miss the pattern
Finder shows where a file lives, but not which development workflow created it or whether it can be regenerated. Tool data may be split across Caches, Application Support, Logs, saved state, and project folders. Hidden directories make the picture even harder to assemble manually.
A useful analysis groups known data by tool, separates project output from source, and connects storage findings to live system pressure. That context matters more than a raw largest-folder list.
A safe maintenance routine
- Close or pause the tools you intend to maintain
- Scan first and review exact paths, categories, and recoverable space
- Prefer re-creatable caches and logs before deleting workspace or project data
- Keep source in version control and maintain a current backup
- Re-scan after cleanup to verify recovery and watch future growth