Skip to main content
LM Studio is a desktop application that allows you to run large language models locally on your computer with an easy-to-use interface. It’s perfect for users who want local AI with a graphical interface for model management.

​
Overview

  • Type: Local provider
  • Cost: Free
  • API Key Required: No
  • Installation Required: Yes
  • Official Website: https://lmstudio.ai/

​
Prerequisites

1

Download LM Studio

Download and install LM Studio from lmstudio.ai for your operating system (Windows, macOS, or Linux).
2

Download a model

Open LM Studio and use the built-in model browser to download a model. Popular options include:
  • Llama 3.2 - Fast and capable
  • Mistral 7B - Excellent general purpose
  • Phi-3 - Small and efficient
  • Gemma 2 - Google’s open model
3

Load the model

In LM Studio, load the downloaded model by selecting it from your collection.
4

Start the server

Click the “Start Server” button in LM Studio to enable the local API server (default port: 1234).
LM Studio provides a user-friendly interface for downloading, managing, and running models without command-line knowledge.

​
Setup in AI Providers

1

Select LM Studio provider

In the AI Providers settings, click Create AI provider and select LM Studio as the provider type.
2

Configure provider URL

Set the Provider URL to:
http://localhost:1234/v1
(Change the port if you configured LM Studio to use a different one)
3

Select model

Click the refresh button to fetch the currently loaded model. The model name will appear automatically.
4

Test the provider

Click Test to verify the connection is working.
Make sure LM Studio’s server is running before using it with AI Providers. The server must be started manually each time you open LM Studio.
Model FamilySizeRAM RequiredBest For
Llama 3.23B~4GBFast, balanced performance
Phi-3 Mini3.8B~4GBEfficient, great quality
Gemma 22B-9B4-12GBGoogle quality, various sizes
Mistral7B~8GBHigh quality general purpose
Llama 3.18B~10GBExcellent instruction following
Qwen 2.57B~8GBStrong multilingual support
LM Studio shows you model sizes and estimated RAM requirements when browsing models, making it easy to choose one that fits your hardware.

​
Key Features

​
Graphical Model Management

  • Browse and download models from Hugging Face
  • Visual interface for model selection
  • Easy model updates and management
  • Model performance metrics

​
Quantization Options

LM Studio offers models in different quantization levels:
  • Q4 - Smaller file size, faster, slightly lower quality
  • Q5 - Balanced option
  • Q8 - Higher quality, larger file size
  • FP16 - Full precision, best quality, largest size

​
Hardware Acceleration

  • Automatic GPU detection and usage
  • CPU-only mode for systems without compatible GPUs
  • Metal support for Apple Silicon Macs
  • CUDA support for NVIDIA GPUs

​
Chat Interface

Test models directly in LM Studio before using them with Obsidian:
  • Interactive chat interface
  • Prompt templates
  • Parameter tuning
  • Performance monitoring

​
Troubleshooting

​
Server Not Running

If you can’t connect to LM Studio:
  1. Open LM Studio application
  2. Load a model from your collection
  3. Click “Start Server” in the Local Server tab
  4. Verify the port number matches your AI Providers settings

​
Model Not Loading

If the model fails to load:
  1. Check you have enough RAM available
  2. Try a smaller model or lower quantization
  3. Close other applications to free up memory
  4. Check LM Studio logs for error messages

​
Slow Performance

Performance depends on your hardware:
  • With GPU: Much faster, can handle larger models
  • CPU only: Slower, stick to smaller models (3B-7B)
  • RAM: More RAM allows larger models and longer contexts
To improve performance:
  1. Use a smaller model
  2. Enable GPU acceleration if available
  3. Choose a lower quantization (Q4 instead of Q8)
  4. Reduce context length settings

​
Connection Refused

If AI Providers can’t connect:
  1. Verify LM Studio server is running (green indicator)
  2. Check the port number in both LM Studio and AI Providers
  3. Ensure no firewall is blocking localhost connections
  4. Try visiting http://localhost:1234 in your browser

​
Advanced Configuration

​
Custom Port

To use a different port:
  1. In LM Studio, go to Settings → Server
  2. Change the port number
  3. Update the Provider URL in AI Providers accordingly

​
Model Parameters

Adjust in LM Studio’s interface:
  • Context Length - How much text the model can process
  • Temperature - Control randomness (0.0-2.0)
  • Top P - Nucleus sampling parameter
  • Repeat Penalty - Reduce repetition
  • GPU Layers - How much to offload to GPU

​
Prompt Templates

LM Studio supports various prompt templates:
  • ChatML
  • Llama
  • Alpaca
  • Vicuna
  • Custom formats
The correct template is usually auto-detected based on the model.

​
Best Practices

  1. Match model to hardware: Choose model size based on your RAM
  2. Use GPU if available: Dramatically faster than CPU-only
  3. Start small: Test with smaller models first
  4. Keep server running: More convenient than starting/stopping
  5. Update models: Check for updated versions periodically
  6. Monitor performance: Use LM Studio’s built-in metrics

​
Comparison with Ollama

LM Studio advantages:
  • Graphical interface
  • Easier for beginners
  • Visual model browsing
  • Built-in chat interface
Ollama advantages:
  • Command-line efficiency
  • Better for automation
  • Lower overhead
  • More scripting options
Both LM Studio and Ollama are excellent choices for local AI. Choose based on your preference for graphical (LM Studio) vs command-line (Ollama) interfaces.

​
Privacy and Offline Use

LM Studio offers complete privacy:
  • All processing happens locally
  • No data sent to external servers
  • Works completely offline
  • No API keys or accounts needed
Perfect for:
  • Sensitive documents
  • Offline environments
  • Privacy-conscious users
  • Air-gapped systems

Build docs developers (and LLMs) love

Get started for freeTalk to us