Skip to main content

Open WebUI Integration

l3mcore is fully compatible with Open WebUI, the most popular web interface for self-hosted AI models.

Configuration​

Option A: Ollama type connection (Recommended)

  1. Open Open WebUI and go to Settings → Connections
  2. In the Ollama API section:
    • Base URL: http://your-lemoe-ip:11435
  3. Click on Save and reload the page. In the model selector you will see all your experts.

Option B: OpenAI type connection

  1. Open Open WebUI and go to Settings → Connections
  2. In the OpenAI API section:
    • Base URL: http://your-lemoe-ip:11435/v1
    • API Key: any value (l3mcore ignores it)
  3. Click on Save and reload the page.

Routing behavior​

Configuration in Open WebUIBehavior in l3mcore
Model: l3mcore Default (or any generic name)The ML router automatically chooses the best expert
Model: programador (exact label)Goes straight to the programador expert, bypassing the router
Model: escritor_creativoGoes straight to the escritor_creativo expert
Use automatic routing

To get the most out of l3mcore, select a generic model and let the router choose. You will see in the logs how it decides in real time.

Example log with Open WebUI​

[API] POST /v1/chat/completions from 192.168.1.50
[API] model=gpt-4, messages=1, stream=true
[Router] Prompt: "Can you write me a poem about the sea?"
[Router] label='escritor_creativo' score=0.89 (embedding match)
[ExpertDispatcher] → api (anthropic/claude-3-5-sonnet)
[Stream] Sending tokens to client...

Troubleshooting​

No models appear in Open WebUI:

  • Verify that l3mcore is running: curl http://your-ip:11435/api/version
  • Check that the Base URL is correct (without /v1 for Ollama, or with /v1 for OpenAI).

Connection error:

  • Make sure the firewall allows port 11435
  • In Docker, use the host IP, not localhost