DEV Community

Elder Fernandes
Elder Fernandes

Posted on • Originally published at selfhoststack-8z4.pages.dev

How to Run Local LLMs with Open WebUI on Docker (Ditch the 0/mo AI Subscriptions)

#opensource #selfhosted #docker #ai

How to Run Local LLMs and Open WebUI on a Cloud VPS (Goodbye $20/mo ChatGPT Plus)

Originally published on SelfHostStack

If you pay $20/month for ChatGPT Plus or Claude Pro, you are paying $240 every single year for:

  1. Rate limits during peak working hours.
  2. Cloud providers retaining your confidential code and business prompts for model training.
  3. Zero control over when models get deprecated, re-aligned, or degraded in coding capabilities.

With modern quantized open-weight models (Llama 3.1, Mistral NeMo, DeepSeek-Coder, and Qwen 2.5), you can host your own private, unrestricted AI workspace on a low-cost VPS with Open WebUI and Ollama.

In this guide, we'll walk through the complete deployment using Docker Compose, setup reverse proxy authentication, and benchmark token generation speeds.

The Economics: SaaS AI vs. Self-Hosted AI

Model / Setup Cost Privacy / Retention Context & Feature Limits
ChatGPT Plus / Claude Pro $20/month ($240/yr) Cloud training, data retained Strict 3-hour message caps, hard rate limits
Open WebUI + Ollama (VPS) ~$6 - $12/month VPS 100% Private (Zero telemetry) Unlimited messages, custom system prompts, RAG document chat
Hybrid (Open WebUI + OpenRouter/DeepSeek API) ~$1 - $3/month (Pay per token) Zero retention API options Access to 100+ models in a unified ChatGPT-like UI

Why Open WebUI is the Superior ChatGPT Interface

Open WebUI (70k+ GitHub stars) is arguably the most polished open-source AI frontend available:

  • Full feature parity with ChatGPT: Chat history, markdown code highlighting, branch editing, and web search.
  • Multi-Model Orchestration: Switch between Ollama local models and external APIs (OpenAI, Anthropic, DeepSeek, Groq) seamlessly.
  • Built-in RAG (Retrieval-Augmented Generation): Upload PDFs, markdown files, and codebases to chat with your private documentation offline.
  • Granular Multi-User RBAC: Invite your whole team without paying per-seat SaaS license fees.

1. The Production Docker Compose Stack

Here is the exact production-ready docker-compose.yml to run Open WebUI paired with Ollama:

version: '3.8'
services:
  open-webui:
    image: ghcr.io/open-webui/open-webui:main
    container_name: open-webui
    restart: always
    ports:
      - "3000:8080"
    environment:
      - OLLAMA_BASE_URL=http://ollama:11434
      - WEBUI_SECRET_KEY=generate_random_secret_string_here
      - ENABLE_SIGNUP=false # Set to false after creating your admin account
    volumes:
      - webui-data:/app/backend/data
    depends_on:
      - ollama

  ollama:
    image: ollama/ollama:latest
    container_name: ollama
    restart: always
    ports:
      - "11434:11434"
    volumes:
      - ollama-models:/root/.ollama

volumes:
  webui-data:
  ollama-models:
Enter fullscreen mode Exit fullscreen mode

2. Step-by-Step Server Setup

Step 1: Provision a High-Compute VPS

For lightweight models (like llama3.2:3b or qwen2.5-coder:1.5b), a standard 4GB RAM cloud instance works great. For larger 8B parameter models, choose a VPS with 8GB RAM and fast NVMe storage:

Step 2: Install Docker & Spin Up the Stack

SSH into your server and run:

# Install Docker Engine
curl -fsSL https://get.docker.com | sh

# Create directory and start stack
mkdir -p ~/ai-stack && cd ~/ai-stack
nano docker-compose.yml # (paste the compose YAML above)
docker compose up -d
Enter fullscreen mode Exit fullscreen mode

Step 3: Pull High-Efficiency Models

Pull lightweight, high-performance models directly inside the Ollama container:

# Ultra-fast coding model (Qwen 2.5 Coder 7B)
docker exec -it ollama ollama pull qwen2.5-coder:7b

# General reasoning model (Llama 3.1 8B)
docker exec -it ollama ollama pull llama3.1:8b

# Extremely fast 3B model for low-resource VPS
docker exec -it ollama ollama pull llama3.2:3b
Enter fullscreen mode Exit fullscreen mode

Step 4: Secure with SSL Reverse Proxy

Use Caddy for automatic HTTPS:

ai.yourdomain.com {
    reverse_proxy localhost:3000
}
Enter fullscreen mode Exit fullscreen mode

3. Hybrid Power: Adding Pay-As-You-Go API Keys

If you need occasional access to frontier models (Claude 3.5 Sonnet, GPT-4o, or DeepSeek-V3) for massive refactoring tasks, you don't need a $20/month subscription:

  1. Open WebUI Settings → Connections → OpenAI API.
  2. Add your OpenRouter or DeepSeek API key.
  3. Now you get local models for everyday questions ($0 cost) and frontier APIs for $0.002 per prompt, saving ~90% of your AI budget.

Explore More Self-Hosted Stacks

Looking to replace other expensive developer and productivity tools? Explore our curated guides with verified Docker templates:

Top comments (0)