Build Your Own Private ChatGPT in 30 Minutes

Emad Mokhtar

15 Sep 2025 — 4 min read

What you'll build

In just 30 minutes, you'll have your private AI assistant running locally on your Mac. No data leaves your machine, and you'll have unlimited conversations without any subscription fees.

Prerequisites

macOS (Big Sur or newer)
Docker Desktop installed and running
LMStudio downloaded from https://lmstudio.ai/download
(Optional) An Apple Silicon Mac for best performance with MLX models

What You'll Get

Complete privacy - your conversations never leave your machine
Offline AI capability - works without internet
Free unlimited usage - no API costs or rate limits
ChatGPT-like interface with conversation history

Steps

Step 1: Install Docker or any container runtime.

You can follow the official document to install Docker. After installing Docker, make sure it's running. Then open the Docker Desktop application and go to Settings -> Resources -> Network. Check the Enable host networking checkbox.

Host networking allows containers that are started with --net=host use localhost to connect to TCP and UDP services on the host. It will automatically allow software on the host to use localhost to connect to TCP and UDP services in the container.

Step 2: Install LMStudio

LMStudio is perfect for this setup because it natively supports MLX models and makes hosting local LLMs straightforward.

MLX is Apple's machine learning framework optimized for Apple Silicon chips, providing significantly faster performance than standard models.

You can use Ollama instead of LMStudio, but some steps may differ and aren't covered in this tutorial.

After installing LMStudio, you need to configure a few settings.

Model Location (optional)

If you want to download models in a different directory or drive. Click on the folder icon and select the path where you want the model downloaded into. I usually download these models on a fast external SSD so it won't fill the OS drive.

Downloading Models

After installing LMStudio, open it and download the MLX model. Click on the magnifier icon.

Uncheck "GGUF" to show only MLX models (which are optimized for Mac), then search for the model you would like to download. I prefer the gpt-oss model.

Run Web Service

LMStudio has its web service. It's not running by default, so you need to run it manually the first time.

Copy the web service's URL - you'll need it in Step 4 to connect OpenWebUI to LMStudio.

Step 3: Install and run OpenWebUI

You'll install OpenWebUI using Docker Compose, which makes updates easier and isolates dependencies from your host machine.

Docker Compose File

openwebui.yaml

services:
  openwebui:
    image: ghcr.io/open-webui/open-webui:main
    restart: always
    volumes:
      - type: bind
        source: ~/.configs/openwebui
        target: /app/backend/data
    network_mode: "host"
  watchtower:
    image: containrrr/watchtower
    restart: always
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
    command: --interval 300 open-webui
    depends_on:
      - openwebui

This Docker Compose file defines two services, one for the OpenWebUI and the other one for Watchtower:

The watchtower is checking for a new image for open-webui. It makes sure the latest container is running.
The openwebui service is running the open-webui image with two configurations:
1. A volume bound to a local path. This makes the container stores your configurations locally. In case you want to upgrade or remove the container, the configuration stays local.
2. Network mode is host so the container will use the host network. This configuration is helpful if you would like to access the OpenWebUI from the internet.

Step 4: Configure LMStudio and OpenWebUI

After installing and running the OpenWebUI, open it in a browser http://localhost:3000. Create your admin user, then go to the Admin Panel.

Then go to Settings -> Connections. Enable the OpenAI API and fill in the LMStudio's web service URL. Please don't forget to add /v1 at the end of the URL.