Turn LM Studio into a Local OpenAI‑Compatible API Server

Overview

LM Studio can expose a local HTTP endpoint that mimics the OpenAI API, allowing client applications to call a loaded model through standard REST calls.

1. Select a Model

Open the Developer tab in LM Studio and choose the desired model from the list (e.g., llama-4-maverick-17b-128e-instruct).

2. Configure Port Exposure

In the Server settings set the listening port (default 1234) and enable CORS so that web pages or other tools can connect.

3. Start the Service

Switch to the Status tab. LM Studio will start an HTTP server and print the available endpoints.

2025-04-26 20:55:13  [INFO] [LM STUDIO SERVER] Success! HTTP server listening on port 1234
2025-04-26 20:55:13  [INFO] [LM STUDIO SERVER] Supported endpoints:
GET http://localhost:1234/v1/models
POST http://localhost:1234/v1/chat/completions
POST http://localhost:1234/v1/completions
POST http://localhost:1234/v1/embeddings
[LM STUDIO SERVER] Logs are saved into /Users/javaedge/.lmstudio/server-logs
Server started.

4. Quick‑Start API Calls

4.1 List Loaded Models

Verify that the server is reachable and see which models are loaded:

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

4.2 Chat Completion

Send a /v1/chat/completions request that follows the OpenAI JSON schema. The request is stateless; the client must include the full conversation history each time.

curl http://127.0.0.1:1234/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "llama-4-maverick-17b-128e-instruct",
    "messages": [
      { "role": "system", "content": "Always answer in rhymes." },
      { "role": "user", "content": "Introduce yourself." }
    ],
    "temperature": 0.7,
    "max_tokens": -1,
    "stream": true
  }'

4.3 Streaming vs. Full Response

If "stream": true, LM Studio returns each generated token as soon as it is produced, enabling low‑latency, incremental display. Setting "stream": false makes the server accumulate the entire completion before sending the response, which can increase latency for long outputs.