lmstudio-typescript

name	lmstudio-typescript
description	LM Studio TypeScript SDK (@lmstudio/sdk), lmstudio-js, local LLM inference, chat completions, structured output, tool use, agentic .act(), plugins, embeddings, model loading, LM Link, lms CLI, llmster headless daemon

LM Studio TypeScript SDK

Package: @lmstudio/sdk (npm). Connects to a running LM Studio instance (desktop app or llmster daemon) over local WebSocket.

Install

npm install @lmstudio/sdk

Quickstart

import { LMStudioClient } from "@lmstudio/sdk";
const client = new LMStudioClient();
const model = await client.llm.model("qwen/qwen3-4b-2507");
const result = await model.respond("Hello");
console.info(result.content);

Reference routing

• Chat completions, streaming, multi-turn → references/chat-completions.md
• Structured output (zod / JSON schema) → references/structured-output.md
• Image/vision input → references/image-input.md
• Tool definition & agentic .act() → references/agentic-tools.md
• Plugins (tools provider, preprocessor, generator) → references/plugins.md
• Embeddings → references/embeddings.md
• Model management (load/unload/list/info) → references/model-management.md
• Configuration (inference & load params) → references/configuration.md
• Infrastructure (headless, LM Link, CLI) → references/infrastructure.md

Key conventions

• client.llm.model() — get any loaded model; client.llm.model("key") — JIT-load specific model
• client.llm.load("key") — force-load a new instance
• .respond() returns an async iterable (streaming) or awaitable (non-streaming via .result())
• Auth: set LM_API_TOKEN env var or pass apiToken to constructor
• Project scaffold: lms create node-typescript