AI with Open WebUI + LiteLLM: Self-Hosted AI Solution

Run OpenWebUI on your own VPS with Hostinger and create a powerful self-hosted AI environment. This setup allows you to use multiple AI models simultaneously, creating a one-to-many scenario where you can compare responses across different models. For example, you could ask what your country would have been like if some historical event didn't happen, and get multiple AI perspectives at once.

One significant advantage of this approach is cost efficiency. Purchasing API tokens can sometimes be more economical than paying standard monthly subscription fees, depending on your usage patterns.

How to Access Multiple AI Models from One Self-Hosted Interface

Step 1: Choose Your Hosting Method

Cloud Option: Set up a VPS (Virtual Private Server) on a hosting provider
- Recommended: KVM2 plan with 8GB RAM and NVMe storage
- Consider a 12-month term for better pricing
On-Premises Option: Host on your own hardware (laptop, NAS, Raspberry Pi)

Step 2: Set Up the Server with OpenWebUI

If using a cloud provider like Hostinger.com:
- Select "Applications" during setup
- Choose Llama as application (includes OpenWebUI)
- Set a secure root password
- Wait for server deployment (usually a few minutes)
Once deployed:
- Access the server management page
- Click "Manage" then "Manage App"
- This opens OpenWebUI running on port 8080

Step 3: Create Your Admin Account

Click "Get Started" on the OpenWebUI welcome page
Create your admin account with full privileges

Step 4: Connect to AI Providers via API Keys

For OpenAI (ChatGPT):
- Go to openai.com/api and sign up/sign in
- Add a payment method (you'll only be charged for what you use)
- Add $5 credit to start
- Go to Settings → API Keys → Create new secret key
- Copy the API key
- In OpenWebUI, go to Profile → Admin Panel → Settings → Connections
- Paste your OpenAI API key and save
For more AI providers (Claude, Gemini, Grok, etc.):
- Install LiteLLM as a proxy server
- SSH into your server or use the browser terminal
- Run: git clone https://github.com/BerriAI/litellm.git
- Navigate to the folder: cd litellm
- Create an env file: nano .env
- Add two configuration lines:
```
LITELLM_MASTER_KEY="sk-randomcharacters"
LITELLM_SALT_KEY="sk-randomcharacters"
```
- Save the file (Ctrl+X, Y, Enter)
- Run: docker-compose up -d
- Access LiteLLM interface at your-ip-address:4000
- Log in with username "admin" and the master key you created

Step 5: Add Models to LiteLLM

In LiteLLM admin panel, go to "Models"
Click "Add Model"
Select provider (Anthropic, Google, OpenAI, etc.)
Choose specific models you want to use
Enter your API key for that provider
Click "Add Model"

Step 6: Create Virtual API Keys

In LiteLLM, go to "Virtual Keys"
Create a new key
Name it (e.g., "kids", "employees")
Select which models this key can access
Set a monthly budget limit if desired
Copy the generated key

Step 7: Connect OpenWebUI to LiteLLM

In OpenWebUI, go to Admin Panel → Settings → Connections
Add your LiteLLM virtual API key under OpenAI API
Set the base URL to: http://localhost:4000 (if hosted on same server)
Click "Verify Connection" then "Save"

Step 8: Create User Accounts & Set Permissions

In OpenWebUI, go to Admin Panel → Settings → Users
Create groups (e.g., "kids", "employees")
Create user accounts and assign them to groups
Set permissions for each group:
- Which models they can access
- Whether they can use tools/plugins
- Access to knowledge bases
Add system prompts to models for safety (e.g., prevent homework cheating)

Monitor Usage and Set Budgets

Check the LiteLLM "Usage" tab to monitor spending
Set budget limits on virtual API keys to prevent overspending
Review conversation logs through the OpenWebUI admin panel

Important Considerations

Be aware of token pricing (pay-per-use model)
Different models have different pricing (GPT-4.5 is much more expensive than GPT-3.5)
Long conversations use more tokens as context is maintained
Monitor usage to avoid unexpected costs