This article was translated by AI

LM Studio

Introduction

LM Studio is a local AI model manager that allows you to run open source large language models such as Mistral, Llama and Phi directly on your computer.

Instead of using cloud APIs, LM Studio runs models locally and exposes a local API endpoint that other applications can connect to.

This setup is useful if you:

Need full control over your data
Work in an offline environment
Must comply with internal security policies
Want to avoid external API costs

Before you begin, make sure LM Studio is installed and running, and that the AI plugin is installed and enabled in ONLYOFFICE.

System requirements

Component	Requirement
OS	Windows (x64, ARM64), macOS (Apple Silicon), Linux (x64)
CPU	AVX2 instruction set support required (x64 systems)
RAM	16 GB recommended
GPU	4 GB VRAM recommended; running on CPU only is possible but significantly slower

Quick setup

Step 1. Install LM Studio

Download LM Studio from the official LM Studio website (available for Windows, macOS and Linux).
Install and launch the application.
Select a local folder where AI models will be stored.

At this stage, LM Studio is installed but not yet connected to ONLYOFFICE.

Step 2. Enable and prepare the local API server

ONLYOFFICE requires two things from LM Studio: a loaded model and a running API server.

Download a model
1. Open the Models section.
2. Choose a model that fits your hardware.
3. Download the model and make sure it is fully loaded.
Start the API server
1. Go to the Server tab.
2. Switch the server status from Stopped to Running or press Cmd+R on macOS.
3. Note the base URL, for example: http://127.0.0.1:1234.

Once the model is downloaded and the server is running, your system functions as a private AI endpoint.

Step 3. Enable the AI plugin

Open ONLYOFFICE Desktop Editors version 9.0.4 or later.
Go to Plugins → Plugin Manager.
Install or update the AI plugin, then close the Plugin Manager.
Go to Plugins → Background plugins and enable the AI plugin.

Step 4. Connect LM Studio to ONLYOFFICE

Open the AI tab → Settings → Edit AI models and click the + icon to add a new model.
Enter the following:
- Model name: LM Studio
- URL: the base URL displayed in the Server tab in LM Studio, for example, http://127.0.0.1:1234/v1
When connecting via a web browser, Enable CORS must be turned on in the Server settings. Without it, the browser will block requests to the LM Studio API.
The list of the models is updated automatically. If you want to do it manually, click Update models.
Assign the tasks you want this model to handle, then save.

Connecting LM Studio

If the connection is successful, ONLYOFFICE will route requests to your local model.

Using the AI assistant

Keyboard shortcuts:

Open the AI assistant	Windows / Linux	`Ctrl`+`/`
Open the AI assistant	macOS	`⌘`+`/`
Clear conversation history	Windows / Linux	`Ctrl`+`Alt`+`/`
Clear conversation history	macOS	`⌘`+`⌥`+`/`
Close the AI agent panel	All platforms	`Esc`

Troubleshooting

Expand all

Collapse all

Cannot connect to LM Studio: Verify that the server is running in the Server tab and that the URL in ONLYOFFICE matches exactly what LM Studio displays, including the port number. If the connection still fails, check whether a firewall or antivirus application is blocking localhost traffic. Restart both applications after making any changes.

Responses are slow: Slow responses are almost always a hardware limitation, not a configuration problem. Larger models consume more RAM and benefit from GPU acceleration. Switching to a smaller quantized model and closing other memory-intensive applications will generally improve response times.

The AI assistant panel opens but does not respond: If the panel opens but no answer is generated, the most common cause is that no model is loaded. The API server can run without a model, but it cannot process requests in that state. Confirm a model is downloaded and actively loaded in LM Studio. Also ensure that the model is assigned to the Chat task in AI settings in ONLYOFFICE.

Model list does not update: The API server must be running before ONLYOFFICE is opened. If you started LM Studio after opening the editor, restart ONLYOFFICE and refresh the model list.