Where Does Text Generation WebUI Store Models and Chats?
Text Generation WebUI keeps models, LoRAs, characters and presets under user_data in its install folder. Layout and overrides.
Last updated
The WebUI stores everything relative to its install folder under user_data. Models go in user_data/models (multi-file models get their own subfolder), LoRAs in loras, characters in characters, presets in presets, logs and cache beside them. The Model tab downloads straight into these spots, so disk planning starts here.
Two flags relocate the tree: --user-data-dir moves everything, --model-dir and --lora-dir move just those parts. Symlinked model folders work for sharing one copy between frontends. Back up loras and characters first: both are irreplaceable, while base models can always be redownloaded.
Where Text Generation WebUI stores this, by platform
<install folder>\user_data\models
Default model home; single GGUF files sit loose, multi-file models in subfolders. Override with --model-dir. The whole tree moves with --user-data-dir.
<install folder>/user_data/models
Same relative layout as Windows. Characters, presets and settings.yaml travel with it. Docker bind-mounts split these into ./models, ./loras, ./characters and ./presets.
<install folder>/user_data/models
Mirrors macOS. Subdirectory loading plus symlinks let one model copy serve several UIs. Training output lands in loras, never in models.
Frequently asked questions
Where do I put a downloaded GGUF model?
Drop single-file GGUFs directly into user_data/models. Multi-file models (safetensors plus tokenizer) belong in their own subfolder inside it. The UI detects both automatically on refresh.
How do I share models between two AI frontends?
Keep one canonical models folder and symlink it into each frontend's expected spot, or point --model-dir at it. Symlinks inside user_data/models to other locations work too. Never duplicate multi-gigabyte weights to save space.
Notice an outdated path? Let us know.