LM Studio Models API

openapi: 3.1.0
info:
  title: LM Studio Local Server Anthropic Compatibility Models API
  version: 1.0.0-beta
  description: 'LM Studio runs large language models locally and exposes a local HTTP server with three families of endpoints: the native LM Studio REST API (beta) under `/api/v1`, an OpenAI-compatible surface under `/v1`, and an Anthropic-compatible surface (`/v1/messages`). The server runs on the developer''s own machine (default `http://localhost:1234`). The native `/api/v1` API adds LM Studio-specific capabilities such as MCP via API, stateful chats, model download/load/unload, generation stats (tokens-per-second, time-to-first-token), and optional API-token authentication. Endpoints and paths are transcribed from the public LM Studio developer documentation; request/response bodies follow the documented OpenAI-style message shapes.'
  contact:
    name: LM Studio
    url: https://lmstudio.ai/docs
  license:
    name: Proprietary (LM Studio application)
    url: https://lmstudio.ai/terms
  x-provenance:
    method: generated
    source: https://lmstudio.ai/docs/app/api/endpoints/rest
    generated: '2026-07-20'
servers:
- url: http://localhost:1234
  description: Default local LM Studio server (bind host/port configurable)
tags:
- name: Models
  description: List, load, unload, and download local models.
paths:
  /api/v1/models:
    get:
      tags:
      - Models
      operationId: listModels
      summary: List loaded and downloaded models
      responses:
        '200':
          description: List of models known to this LM Studio instance.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ModelList'
        '401':
          $ref: '#/components/responses/Unauthorized'
  /api/v1/models/load:
    post:
      tags:
      - Models
      operationId: loadModel
      summary: Load a model into memory
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ModelLoadRequest'
      responses:
        '200':
          description: Model load acknowledged.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/Model'
        '400':
          $ref: '#/components/responses/BadRequest'
        '404':
          $ref: '#/components/responses/NotFound'
  /api/v1/models/unload:
    post:
      tags:
      - Models
      operationId: unloadModel
      summary: Unload a model from memory
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ModelUnloadRequest'
      responses:
        '200':
          description: Model unload acknowledged.
        '404':
          $ref: '#/components/responses/NotFound'
  /api/v1/models/download:
    post:
      tags:
      - Models
      operationId: downloadModel
      summary: Download a model from the LM Studio / Hugging Face catalog
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/ModelDownloadRequest'
      responses:
        '200':
          description: Download started.
        '400':
          $ref: '#/components/responses/BadRequest'
  /api/v1/models/download/status:
    get:
      tags:
      - Models
      operationId: getDownloadStatus
      summary: Check model download progress
      responses:
        '200':
          description: Current download progress.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/DownloadStatus'
components:
  schemas:
    ModelUnloadRequest:
      type: object
      properties:
        model:
          type: string
      required:
      - model
    ModelList:
      type: object
      properties:
        object:
          type: string
          example: list
        data:
          type: array
          items:
            $ref: '#/components/schemas/Model'
    DownloadStatus:
      type: object
      properties:
        model:
          type: string
        progress:
          type: number
        state:
          type: string
    Error:
      type: object
      properties:
        error:
          type: object
          properties:
            message:
              type: string
            type:
              type: string
            code:
              type: string
    ModelLoadRequest:
      type: object
      properties:
        model:
          type: string
          description: Model key/identifier to load.
        config:
          type: object
          description: Optional load configuration (context length, GPU offload).
      required:
      - model
    Model:
      type: object
      properties:
        id:
          type: string
        object:
          type: string
        type:
          type: string
          enum:
          - llm
          - embeddings
          - vlm
        arch:
          type: string
        state:
          type: string
          enum:
          - loaded
          - not-loaded
        max_context_length:
          type: integer
    ModelDownloadRequest:
      type: object
      properties:
        model:
          type: string
          description: Model identifier / Hugging Face repo to download.
      required:
      - model
  responses:
    NotFound:
      description: The referenced model or resource does not exist.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    Unauthorized:
      description: A required API token was missing or invalid.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
    BadRequest:
      description: The request was malformed or referenced an unavailable model.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
  securitySchemes:
    ApiToken:
      type: http
      scheme: bearer
      description: Optional API token. The native `/api/v1` server can be configured to require a bearer token; by default the local server accepts unauthenticated requests from the loopback interface.