"I keep a Samsung T7 with six GGUF models on it and carry it between my home workstation and office desktop. Plugging in and updating the lm studio portable model path takes about twenty seconds. Both machines have their own LM Studio install; the models travel without re-downloading. Saved me roughly 80 GB of duplicated storage."
LM Studio portable: running the application from external storage
How to configure lm studio portable mode — storing models on a USB drive or external SSD, setting a custom model directory, and understanding the performance trade-offs between USB 3.x and Thunderbolt connections.
Vital Points
LM Studio portable usage means pointing the model directory at an external SSD. The app itself stays on the host machine while models live on removable storage. USB 3.2 Gen 2 (10 Gbps) is the practical minimum for comfortable load times; Thunderbolt 3/4 is the best option. FAT32 and exFAT work on Windows; ext4 is better on Linux. Model load times are longer than from internal NVMe but perfectly workable on fast USB connections.
What "portable" means for LM Studio
LM Studio portable usage is primarily about relocating the model directory — the application itself runs from a standard install on the host machine.
LM Studio does not ship a single self-contained executable in the sense that the entire application state can live on a thumb drive and run on any machine without installation — the app binary needs to be installed on the host OS per-platform. What is portable in practice is the model library: by pointing the model directory to an external drive, you carry your full collection of GGUF model files between machines without re-downloading them. Plug in the drive, update the model directory path in Settings, and LM Studio on that host sees your complete library.
On Linux, the AppImage is itself a portable format, so the combination of the AppImage and an external model directory gives you a genuinely cross-machine setup — copy the AppImage and the model folder to a drive, connect it to any Linux machine with FUSE support, and you are running without installation. That is the closest lm studio portable gets to a fully self-contained experience.
This pattern is common among consultants, researchers who carry a drive between home and office, and developers who maintain a consistent model set across multiple workstations. The workflow is straightforward: one external SSD holds all model files, and each machine's LM Studio installation simply reads from it.
Setting a custom model directory
The model directory setting in LM Studio is where all the portable magic happens — one field, one path change, and your full library follows the drive.
To point LM Studio at an external drive, open the application, navigate to Settings (the gear icon or via the menu bar), and find the "Model directory" field. Paste in the full path to the folder on your external drive where models are stored — or use the browse button to select it visually. Click Save. LM Studio immediately scans the new path and populates the model list with whatever GGUF files it finds there.
On Windows, the path will look something like E:\LMStudio\models where E: is the drive letter assigned to your external storage. On macOS, it will be /Volumes/MyDriveName/LMStudio/models. On Linux, the mount point varies by distribution and configuration, but a typical path is /media/username/MyDriveName/LMStudio/models.
LM Studio stores this setting in its configuration file on the host machine. The setting persists between sessions. If the external drive is not connected when LM Studio starts, the model directory path remains set but the library shows empty — reconnect the drive and the models reappear without any further configuration.
USB and Thunderbolt performance notes
Interface speed determines how long you wait for a model to load — the numbers vary dramatically between USB generations and Thunderbolt.
Model loading is a sequential read operation: the GGUF file streams from disk into memory. That means raw interface throughput translates directly into load time. The table below summarises the practical read speeds and load time estimates for a typical 5 GB (7B Q4) model.
USB 2.0 at a theoretical 480 Mbps (60 MB/s real) would take roughly 80 seconds to load a 5 GB model — too slow for iterative work. USB 3.0 at 5 Gbps (400 MB/s real) brings that to around 12 seconds. USB 3.2 Gen 2 at 10 Gbps (900+ MB/s real) gets under 6 seconds, which is comparable to a SATA SSD. Thunderbolt 3 and 4 at 40 Gbps can push 2–3 GB/s in real-world RAID configurations, bringing load times under 2 seconds — essentially matching a midrange NVMe SSD.
The SSD inside the enclosure matters as much as the cable interface. A slow QLC SSD in a Thunderbolt enclosure will not hit Thunderbolt's theoretical ceiling. For best results, pair a fast NVMe SSD with a Thunderbolt enclosure. Samsung T7 Shield and WD My Passport SSD are popular USB 3.2 Gen 2 enclosures that perform well for model storage at a lower cost.
The FTC has published guidance on storage device labeling and performance claims (ftc.gov) that is worth reading when evaluating external drive marketing speeds against real-world throughput.
Filesystem considerations
Filesystem choice affects cross-platform compatibility and maximum file size — relevant when model files regularly exceed 4 GB.
FAT32 has a 4 GB per-file limit, which is a problem: many GGUF model files exceed this threshold. A 13B Q4 model is 8–9 GB per file, and a 70B Q4 model is a single file exceeding 40 GB. Format your external drive with exFAT for cross-platform Windows/macOS use, or ext4 for Linux-primary usage. Both support large files without issue.
On macOS, APFS is an option for drives used solely with Macs. It handles large files natively and has better journaling. On Windows, NTFS is the other option for Windows-only use. If the drive will travel between operating systems, exFAT is the safest choice — it is supported natively on all three platforms without additional drivers.
| Interface | Real read speed (typical) | Usability for model storage |
|---|---|---|
| USB 2.0 | ~40–60 MB/s | Poor — load times 60–90 s for 5 GB model |
| USB 3.0 / 3.2 Gen 1 (5 Gbps) | ~300–400 MB/s | Acceptable — 12–18 s for 5 GB model |
| USB 3.2 Gen 2 (10 Gbps) | ~800–1000 MB/s | Good — 5–7 s for 5 GB model |
| Thunderbolt 3 / 4 (40 Gbps) | ~2000–3000 MB/s | Excellent — 1–2 s for 5 GB model, near-NVMe |
Practitioner note
Frequently asked questions
Common questions about using LM Studio with models stored on external or removable storage.
Yes. Open Settings in LM Studio and update the Model directory path to point to your external drive. LM Studio reads from that location on every launch without copying files to the internal drive. An SSD is strongly recommended — spinning hard drives produce very slow model load times that make iterative work frustrating.
USB 3.2 Gen 2 (10 Gbps) or faster is the practical minimum for comfortable model loading. Thunderbolt 3 and 4 (40 Gbps) produce near-NVMe load speeds. USB 3.0 at 5 Gbps works but extends load times for larger models to 12–18 seconds per 5 GB. USB 2.0 is too slow for anything beyond very small models.
Yes. Place the AppImage on the USB drive, mark it executable, and run it from the mounted path. Make sure the filesystem on the drive supports executable permissions — ext4 works natively; exFAT may require specific mount options to enable execute bits. Models can be stored on the same drive by setting the model directory to a folder on the drive.
Yes. The model directory setting is saved to the LM Studio configuration file on the host machine and persists between launches. If the external drive is not connected when LM Studio starts, the app opens normally but shows an empty model library. Reconnect the drive and the models reappear without any additional configuration steps.