How to Run Local AI LLM Models on Ubuntu with LM Studio (2026 Step-by-Step Guide)

Published on August 15, 2026 • By LMSA
How to Run Local AI LLM Models on Ubuntu with LM Studio (2026 Step-by-Step Guide)

Running large language models (LLMs) locally has become one of the most practical ways to use AI without sending your data to the cloud. If you're on Ubuntu and want a simple, GUI-based way to download, run, and manage open-source LLMs on your own hardware, LM Studio is one of the easiest tools available. This guide walks you through installing LM Studio on Ubuntu, downloading your first model, running it locally, and even accessing it from your Android phone using LMSA, a dedicated mobile client built for exactly this purpose.

What Is LM Studio?

LM Studio is a free, cross-platform desktop application for discovering, downloading, and running open-source LLMs directly on your own computer. It wraps the popular llama.cpp inference engine in a polished, Electron-based graphical interface, so you don't need to touch the command line to load a GGUF model and start chatting. It also integrates with Hugging Face's model library, giving you access to thousands of community models, and it can spin up a local OpenAI-compatible API server so you can plug your own apps into it.

Compared to alternatives like Ollama (which is command-line first) or GPT4All (which is more beginner-oriented but has a smaller model catalog), LM Studio strikes a good balance: it's approachable for newcomers but flexible enough for developers who want fine-grained control over quantization, GPU offloading, and context length.

System Requirements for LM Studio on Ubuntu

Before you install anything, make sure your machine meets these baseline requirements:

  • OS: Ubuntu 22.04 LTS or newer (Ubuntu 24.04 LTS is fully supported)
  • CPU: x86_64 processor with AVX2 instruction set support — this is a hard requirement for most GGUF model backends
  • RAM: 16 GB is strongly recommended. You can technically run small 1B–3B parameter models on 8 GB, but 7B–8B models need 16 GB or more
  • GPU (optional but recommended): An NVIDIA GPU with CUDA support, or an AMD GPU with ROCm, will dramatically speed up inference. A minimum of 6 GB of VRAM is suggested for a meaningful performance boost over CPU-only inference

If you're running a CPU-only setup, you can still run smaller models comfortably — you'll just want to stick with lighter quantizations like Q4_K_M.

Step 1: Download LM Studio for Linux

  1. Go to the official LM Studio download page.
  2. Select Linux as your operating system.
  3. Download the file — it will come as an AppImage, something like LM_Studio-x.x.x.AppImage.

The AppImage format is portable and works across nearly all major Linux distributions without a complex installation process, which is why it's the recommended route for Ubuntu users.

Step 2: Install Required Dependencies

AppImages need FUSE (Filesystem in Userspace) to unpack and run themselves. On newer Ubuntu releases, libfuse2 isn't installed by default, and its absence is the single most common reason LM Studio fails to launch. Install it with:

sudo apt update
sudo apt install libfuse2

You may also need a handful of GUI libraries if your system is missing them:

sudo apt install libatk1.0-0 libatspi2.0-0 libasound2t64

Step 3: Make the AppImage Executable

Open a terminal, navigate to your Downloads folder, and grant the file execute permissions:

cd ~/Downloads
chmod +x LM_Studio-*.AppImage

Step 4: Launch LM Studio

You can now run the app directly:

./LM_Studio-*.AppImage

If you hit a sandbox error — which sometimes happens in containerized or virtualized environments — launch it with the --no-sandbox flag instead:

./LM_Studio-*.AppImage --no-sandbox

Step 5: Add LM Studio to Your Application Menu (Optional)

Running the AppImage from the terminal every time gets old fast. To give LM Studio a proper icon in your Ubuntu application launcher, create a .desktop file:

nano ~/.local/share/applications/LM-Studio.desktop

Paste in the following, adjusting the paths to match where you saved the AppImage:

[Desktop Entry]
Type=Application
Name=LM Studio
Exec=/home/YOUR_USERNAME/Downloads/LM_Studio-x.x.x.AppImage --no-sandbox
Icon=/home/YOUR_USERNAME/Downloads/lmstudio-icon.png
Terminal=false
Categories=Development;AI;

Save with Ctrl+O, exit with Ctrl+X, then refresh the desktop database:

update-desktop-database ~/.local/share/applications

LM Studio will now show up alongside your other applications, ready to launch with a click. Alternatively, you can use a tool like AppImageLauncher to handle this integration automatically.

Step 6: Download Your First Model

Once LM Studio is open:

  1. Click the Model Search (magnifying glass) icon in the sidebar.
  2. Search for a beginner-friendly model — something like Llama 3.2 1B Instruct or Gemma 3 is a good starting point if you're testing on modest hardware.
  3. Choose a quantization level. Q4_K_M is a solid default that balances quality and memory usage; Q8_0 gives higher quality at the cost of more RAM/VRAM.
  4. Click download and wait for it to finish.

If you have an NVIDIA or AMD GPU, go to Settings → Hardware Settings, enable GPU Offload, and push the layer slider up (often toward 99 layers, depending on your VRAM) to get the model running as much as possible on your graphics card instead of the CPU.

Step 7: Chat with Your Local Model

  1. Click Chats in the sidebar.
  2. Start a new chat and select your downloaded model from the dropdown.
  3. Type a message and hit enter.

If you get a coherent response, everything is working — you now have a fully private, offline AI chatbot running entirely on your own Ubuntu machine, with no data leaving your computer.

Step 8: Turn On the Local API Server (Optional, for Developers)

LM Studio can host a local server that mimics the OpenAI API format, which means you can point almost any OpenAI-compatible tool, script, or IDE extension at it.

  1. Click the Local Server icon (the two-way arrow) in the sidebar.
  2. Select your loaded model.
  3. Click Start Server.

By default, this runs at http://127.0.0.1:1234. You can verify it's live with:

curl http://127.0.0.1:1234/v1/models

From here, you can wire the endpoint into Python scripts, VS Code extensions like Continue.dev, or any custom application that speaks the OpenAI chat completions format.

Step 9: Chat With Your Local LLM from Your Android Phone Using LMSA

Once your model is running on your Ubuntu machine, you don't have to stay tied to your desk to use it. LMSA is a dedicated Android app built specifically to connect to LM Studio (and Ollama) servers running on your local network, letting you chat with your own models from your phone or tablet.

Here's what makes LMSA worth installing:

  • 100% private by design — conversations stay on your local network; nothing routes through LMSA's own servers when you're using a local model, and chat history is stored in an encrypted database on your device.
  • In-app model switching — swap between any models loaded in LM Studio directly from your phone, without walking back to your computer.
  • Voice Mode — talk to your local LLM hands-free. Your phone's native speech recognition handles speech-to-text, and Android's built-in text-to-speech reads responses back to you, so no audio ever leaves your device for a third-party voice API.
  • Multimodal and file support — send text and, depending on your model, images and documents for the AI to process.
  • Prompt library, custom personalities, and thinking mode — see full reasoning chains for models like DeepSeek-R1, save reusable prompts, and build custom AI personas.
  • Optional cloud fallback via OpenRouter — if you ever want access to 100+ cloud models alongside your local ones, LMSA supports that too, using your own API key.
  • Free core app — the local server connection and voice mode are free with unlimited local chat completions; a one-time purchase unlocks extras like custom endpoints, biometric lock, and an ad-free experience.

Connecting LMSA to LM Studio in 4 Steps

  1. In LM Studio on your Ubuntu machine, start the Local Server (Step 8 above) and note the port (default 1234).
  2. Make sure your Ubuntu PC and your Android phone are connected to the same Wi-Fi network.
  3. Find your PC's local IP address (run hostname -I in a terminal on Ubuntu).
  4. Open LMSA, go to its server settings, and enter your PC's IP address and port (e.g., 192.168.1.42:1234). Start chatting.

If you ever want to reach your local model while away from home, tools like Tailscale can create a secure tunnel back to your Ubuntu machine, letting you keep the same private setup on the go.

Troubleshooting Common Issues

  • AppImage won't launch: Almost always a missing libfuse2 dependency — reinstall it with the command in Step 2.
  • Sandbox error on launch: Add the --no-sandbox flag when running the AppImage.
  • Slow responses: Check whether GPU offload is enabled in Settings → Hardware Settings, and confirm your GPU drivers are working with nvidia-smi (for NVIDIA cards).
  • Model won't load / out of memory: Try a smaller model or a more aggressive quantization (e.g., Q4_K_M instead of Q8_0).
  • LMSA can't connect: Double-check both devices are on the same Wi-Fi network, confirm the LM Studio server is actually running, and verify you entered the correct local IP and port.

Final Thoughts

LM Studio makes running local LLMs on Ubuntu genuinely approachable — no command-line wrangling required, just an AppImage, a model download, and a chat window. And once your local AI server is up and running, pairing it with LMSA on Android turns your phone into a private, portable interface for the same models, so your AI setup isn't chained to your desk. Together, LM Studio and LMSA give you a complete, privacy-first local AI stack — from your Ubuntu desktop all the way to your pocket.