Open WebUI

Ollama and Slotstream give you a model and a port. Neither gives you somewhere to talk to it, keep the conversation, upload a PDF, or let your family use it. Open WebUI is that missing half: a familiar chat interface in front of local models, hosted APIs, or both at once.

Run it link

  docker run -d -p 3000:8080 \
  --add-host=host.docker.internal:host-gateway \
  -v open-webui:/app/backend/data \
  --name open-webui --restart always \
  ghcr.io/open-webui/open-webui:main
  

Open http://localhost:3000. Variants save you wiring: :ollama bundles Ollama in the same container, :cuda adds GPU acceleration (with --gpus all). If you’d rather not use Docker:

  pip install open-webui
open-webui serve
  

To point it at hosted models instead, pass the key — -e OPENAI_API_KEY=… — or add providers in the settings. It speaks the OpenAI API shape, so anything that does works, including Slotstream on http://localhost:11434/v1.

What it adds over a raw model link

• Switch models mid-conversation, local and hosted side by side, and compare answers from several at once.
• Chat with your documents. Upload files and reference them with #; retrieval is built in rather than an add-on.
• Web search as a toggle, so a local model stops being stuck in its training data.
• Multiple users with roles — the reason this becomes a household or team service rather than a personal one.
• Pipelines and tools for custom Python that runs between the message and the model.

Worth knowing link

• The licence is not plain BSD. It’s BSD-3 with a branding clause: you may not remove or replace the Open WebUI name and logo unless your deployment has 50 or fewer users in a rolling 30 days, you’ve been given written permission, or you hold an enterprise licence. Self-hosting for yourself, a team, or a family is unaffected. White-labelling it inside a product is not.
• The UI is not the bottleneck; the model is. A friendlier front end doesn’t make a 7B model answer like a frontier one — see the honest expectations on the Ollama page.
• If you expose it beyond localhost, put auth in front of it and keep it off the open internet unless you mean it. It holds your conversations and your API keys.

Next link

Local models are one source of tokens; stacked free tiers are another → FreeLLMAPI