LM Studio

Connect hi-shell to LM Studio for local LLM inference with a user-friendly graphical interface. LM Studio makes it easy to download and run models with hardware acceleration.

Prerequisites

• LM Studio installed and running
• At least one model downloaded in LM Studio
• Local API server enabled in LM Studio

Setup

1. Install LM Studio

Download from lmstudio.ai and install.

2. Download a Model

In LM Studio:

1. Search for a model (e.g., “Llama 3.2”)
2. Click “Download”
3. Wait for the download to complete

Recommended models:

• Llama 3.2 1B/3B Instruct — Fast, good quality
• Phi-3 Mini — Compact and capable
• Mistral 7B Instruct — High quality
• Qwen 2.5 Coder — Great for code

3. Enable the Local API Server

In LM Studio:

1. Go to the “Local Server” tab (the double-arrow icon)
2. Select your downloaded model from the dropdown
3. Click “Start Server”
4. The server runs on http://localhost:1234 by default

4. Configure hi-shell

hi-shell --init
# Select "Local" → "LM Studio"
# Confirm the URL (default: http://localhost:1234)
# Enter the model identifier

Manual Configuration

Edit ~/.config/hi-shell/config.toml:

llm_provider = "Local"
local_provider = "LmStudio"
local_url = "http://localhost:1234"
local_model = "lmstudio-community/Llama-3.2-1B-Instruct-GGUF"

Configuration Options

Custom Port

If you changed the LM Studio port:

local_url = "http://localhost:8080"

Verifying the Connection

Test that hi-shell can connect to LM Studio:

hi-shell list all text files

You can also verify LM Studio’s API directly:

curl http://localhost:1234/v1/models

Troubleshooting

“Connection refused” Error

• Make sure LM Studio is running
• Verify the Local Server is started (not just the model loaded)
• Check the correct port (default: 1234)

“Model not found” Error

• Make sure a model is loaded in the Local Server tab
• The model must be actively running (green indicator)

Slow Responses

• Use a smaller quantized model
• Enable GPU acceleration in LM Studio settings
• Close other GPU-intensive applications

LM Studio vs. Ollama