LM Studio is great for running LLMs locally, but by default it puts configs, cache, and models in your %USERPROFILE% folder. If you have a small SSD or want everything on an external drive, here's the clean, official way to make it fully portable.
The Problem
- Models are huge (7B ≈ 4GB, 70B ≈ 40GB+)
- Cache and conversations pile up in
%USERPROFILE% - Moving LM Studio's folder doesn't move the data
The Solution: .lmstudio-home-pointer
LM Studio has a hidden feature — it checks for a file named .lmstudio-home-pointer in your user home. If found, it treats the path inside as its home folder.
Step by Step
| Step | Action |
|---|---|
| 1 | Create file C:\Users\<yourname>\.lmstudio-home-pointer
|
| 2 | Write your desired path inside, e.g. X:\LM Studio
|
| 3 | Install LM Studio for all users to that exact path |
| 4 |
Crucial: The final folder must be named exactly LM Studio
|
| 5 | Launch — all data now lives there |
Resulting Structure
X:\LM Studio
├── .internal\ # App state, cache, downloads
├── bin\lms.exe # Official CLI
├── conversations\ # Chat history
├── extensions\ # Backends: CPU, CUDA 11/12, Vulkan
├── hub\ # Models from LM Studio Hub
├── models\ # Your GGUF files
└── LM Studio.exe # Main app
plain
Migrating to Another PC
- Copy the entire
LM Studiofolder - On the new PC, create
.lmstudio-home-pointerpointing to it - Done — all models, settings, and conversations come with you
Verified Setup
| Component | Detail |
|---|---|
| OS | Windows 11 23H2 |
| GPU | NVIDIA RTX 3070 Ti 8GB |
| LM Studio | v0.3.x |
| Backends tested | CUDA 12, Vulkan, CPU |
Full Guide
All details: GitHub Gist
Have you tried this? Any quirks on your setup? Let me know in the comments!
Top comments (1)
Portable LM Studio is a practical angle because the hard part is not only running the model, it is making the environment repeatable. I would document the model cache path and GPU/runtime assumptions as carefully as the app folder.