Open-WebUI: Self-Hosted ChatGPT UI + Ollama

Open-WebUI is an open-source ChatGPT clone that runs locally and connects to Ollama, OpenAI, or other LLM backends. This guide covers installation with Ollama and fixes for common GitHub issues.

Quick Overview

• Open-WebUI: ChatGPT-like web interface, self-hosted
• Ollama: Local LLM runtime (Llama 3, Mistral, etc.)
• Requirements: Docker + 8GB+ RAM (16GB for 7B models)
• Benefits: Privacy, no API costs, full control

Quick Start with Docker

Fastest way to get running - Docker Compose with both Open-WebUI and Ollama:

# Create docker-compose.yml
cat > docker-compose.yml << 'EOF'
version: "3.8"

services:
  open-webui:
    image: ghcr.io/open-webui/open-webui:main
    container_name: open-webui
    ports:
      - "3000:8080"
    environment:
      - OLLAMA_BASE_URL=http://ollama:11434
      - WEBUI_SECRET_KEY=your-secret-key-here
    volumes:
      - ./open-webui/data:/app/backend/data
    restart: unless-stopped

  ollama:
    image: ollama/ollama:latest
    container_name: ollama
    ports:
      - "11434:11434"
    volumes:
      - ./ollama/data:/root/.ollama
    restart: unless-stopped
EOF

# Start everything
docker-compose up -d

# Pull a model (optional, downloads on first use)
docker exec -it ollama ollama pull llama3

# Open http://localhost:3000

Connecting Ollama Models

Once running, Open-WebUI auto-discovers Ollama. Here's how to pull and use models:

# List available models
docker exec -it ollama ollama list

# Pull popular models
docker exec -it ollama ollama pull llama3          # 4.7GB - Meta's Llama 3 8B
docker exec -it ollama ollama pull mistral         # 4.1GB - Mistral 7B
docker exec -it ollama ollama pull codellama       # 3.8GB - Code specialist
docker exec -it ollama ollama pull phi3            # 2.2GB - Microsoft's 3.8B
docker exec -it ollama ollama pull gemma:2b        # 1.6GB - Google's Gemma 2B

# Refresh Open-WebUI to see new models
# Click "Settings" → "Models" → Select from dropdown

Common Problems & Solutions

Issue #1123: "Connection Refused" to Ollama

github.com/open-webui/open-webui/issues/1123

Problem: Open-WebUI shows "Error connecting to Ollama" even though Ollama is running.

What I Tried: Restarting containers, checking ports - Ollama worked via CLI but not from Open-WebUI.

Actual Fix: Docker networking issue. When both run in Docker, localhost doesn't work. Use the service name:

# WRONG:
environment:
  - OLLAMA_BASE_URL=http://localhost:11434

# RIGHT:
environment:
  - OLLAMA_BASE_URL=http://ollama:11434

# Or if Ollama is on host (not in Docker):
environment:
  - OLLAMA_BASE_URL=http://host.docker.internal:11434

# Add extra_hosts for host.docker.internal:
extra_hosts:
  - "host.docker.internal:host-gateway"

The ollama in the URL matches the service name in docker-compose.yml, not localhost.

Issue #987: Models Not Showing in Dropdown

github.com/open-webui/open-webui/issues/987

Problem: Pulled models in Ollama, but Open-WebUI shows "No models available" in the dropdown.

What I Tried: Refreshing page, restarting Open-WebUI - models still missing.

Actual Fix: Open-WebUI caches the model list. Need to trigger a refresh via API or manual config:

# Solution 1: Trigger refresh via API
curl -X POST http://localhost:3000/api/config/updates/models \
  -H "Content-Type: application/json"

# Solution 2: Restart Open-WebUI after pulling models
docker-compose restart open-webui

# Solution 3: Add models manually in Settings
# Go to Settings → Providers → Ollama → Add Model
# Enter: "llama3", "mistral", etc.

# Solution 4: Check Ollama is actually reachable
docker exec -it open-webui curl http://ollama:11434/api/tags

# Should return JSON with models list

Issue #1456: Authentication Required After Update

github.com/open-webui/open-webui/issues/1456

Problem: Updated Open-WebUI, now it asks for login but never created an admin password. Can't access the UI.

What I Tried: Reinstalling, clearing data - kept asking for authentication.

Actual Fix: Newer versions enable auth by default. Need to create first user via CLI or disable auth:

# Option 1: Disable authentication (development only)
environment:
  - WEBUI_AUTH=false  # Disables login requirement

# Option 2: Create admin user via CLI
docker exec -it open-webui python -c "
from backend.models import User
from backend.database import DB
db = DB()
user = User(
    username='admin',
    email='admin@example.com',
    password='your-password',
    role='admin',
    is_active=True
)
db.add(user)
db.commit()
"

# Option 3: Reset auth and create new user
# Stop container
docker-compose down

# Remove auth database
rm -rf ./open-webui/data/webui.db

# Start and create new account at first visit
docker-compose up -d

Issue #1324: Slow Response Times (10+ seconds)

github.com/open-webui/open-webui/issues/1324

Problem: Responses take 10-20 seconds even for simple queries. CPU usage is low, not a hardware bottleneck.

What I Tried: Switching models, increasing Ollama memory - minimal improvement.

Actual Fix: Two issues: Ollama's default context window is too large, and Open-WebUI wasn't using GPU acceleration:

# 1. Enable GPU in Ollama (if available)
docker exec -it ollama ollama run llama3 --gpu

# 2. Reduce context size for faster responses
# In Open-WebUI: Settings → Models → llama3 → Parameters
# Set "Max Tokens" to 2048 (default is 4096)

# 3. Enable streaming
environment:
  - ENABLE_STREAMING=true

# 4. Use faster quantization
docker exec -it ollama ollama pull llama3:instruct-q4_K_M
# q4_K_M is faster than default q8_0 with minimal quality loss

# 5. Adjust Ollama num_thread
docker exec -it ollama ollama run llama3 --num-thread 8
# Set to your CPU core count

With these changes, response time dropped from 15s to 3-4s on my system.

Issue #1678: Chat History Not Persisting

github.com/open-webui/open-webui/issues/1678

Problem: Chat history disappears when refreshing page or restarting container.

What I Tried: Checking volume mounts - they looked correct but data wasn't persisting.

Actual Fix: The data directory path changed in newer versions. Update volume mount and check permissions:

# OLD path (doesn't work in 0.3.x+):
volumes:
  - ./open-webui:/app/backend/data

# NEW path (0.3.x+):
volumes:
  - ./open-webui/data:/app/backend/data

# Also ensure directory exists with correct permissions:
mkdir -p ./open-webui/data
chmod 777 ./open-webui/data

# Check inside container:
docker exec -it open-webui ls -la /app/backend/data

# Should show webui.db file

If still losing data, check if SQLite database is being written to /tmp instead of mounted volume.

Connecting OpenAI API

You can mix local models with OpenAI for GPT-4 access when needed:

# Add OpenAI as fallback provider
environment:
  - OPENAI_API_KEY=sk-your-key-here
  - OPENAI_API_BASE=https://api.openai.com/v1

  # Or use a proxy:
  - OPENAI_API_BASE=https://your-proxy.com/v1

In the UI, select "OpenAI" from the model dropdown to use GPT-4, or "Ollama" for local models.

Model Selection Guide

Model	Size	Best For	RAM
llama3	4.7GB	General purpose	8GB
mistral	4.1GB	Chat, reasoning	8GB
codellama	3.8GB	Code generation	8GB
phi3	2.2GB	Fast, lightweight	4GB
gemma:2b	1.6GB	Smallest viable	3GB

Performance Tips

GPU Acceleration

# Enable GPU in Ollama
services:
  ollama:
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

Quantization Trade-offs

# Smaller quantization = faster, less RAM, more quality loss
docker exec -it ollama ollama pull llama3:q4_K_M    # Fast, good quality
docker exec -it ollama ollama pull llama3:q8_0      # Slower, best quality
docker exec -it ollama ollama pull llama3:q2_K      # Fastest, most quality loss

Reverse Proxy with HTTPS

# nginx.conf
server {
    listen 443 ssl;
    server_name chat.yourdomain.com;

    ssl_certificate /etc/nginx/ssl/cert.pem;
    ssl_certificate_key /etc/nginx/ssl/key.pem;

    location / {
        proxy_pass http://localhost:3000;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";
        proxy_set_header Host $host;
        proxy_cache_bypass $http_upgrade;
    }
}