Integration guide
LM Studio Local Server Integration
LM Studio's local server exposes a desktop-run model endpoint for apps and prototypes that need local inference.
Recommended path
LM Studio Local Server local AI quick start
- 1. Download and load the model in LM StudioChoose a model that fits the machine, load it successfully, and test one prompt in LM Studio before exposing the server to another application.
- 2. Start the local serverEnable the server from the Developer tab or run lms server start --port 1234, then verify /v1/models from the client machine.
- 3. Point the client at /v1 and secure wider accessSet the OpenAI-compatible base URL to http://localhost:1234/v1, test the exact endpoint, and enable authentication before listening beyond a trusted local machine.
Best for
Builders who want to test models locally, then point tools at a local server where supported.
Model support
LM Studio runs an OpenAI-compatible server over the models you have loaded, exposing /v1/models, /v1/chat/completions, /v1/completions, /v1/embeddings and /v1/responses. Existing OpenAI clients work by changing the base URL.
Choose models by role
Chat and application responses
Gemma 3 27B
Use an instruction model that fits local memory and supports the client's required context and features. LM Studio serves only a model that has been downloaded and loaded.
Open the model profile →Embeddings
Qwen3 Embedding
Load a dedicated embedding model when the application calls /v1/embeddings. Do not assume the chat model is also the right retrieval model.
Open the model profile →How to use this integration
- Serve a model to other applications on your machine through a familiar OpenAI-shaped API
- Browse, download and swap models through a GUI before exposing them to an application
- Run a headless server on a workstation and reach it from a laptop over the LAN
- Test how an application behaves against a local model without changing its client library
Connecting a local model
Start the server from the Developer tab or with the CLI, lms server start --port 1234, then point clients at http://localhost:1234/v1. A headless build is available for machines with no desktop session.
Tradeoffs
The weights are local but the runner is not open: LM Studio is proprietary software licensed for personal and internal business use, with its source treated as a trade secret and some features offered for a fee. On an open-source-first stack that is the trade to weigh against its convenience. Unlike Ollama, it publishes no per-endpoint list of unsupported OpenAI parameters, so compatibility gaps have to be found by testing.