Integration guide
Open WebUI Integration
Self-hosted chat workspace that connects users to local and hosted model backends.
Recommended path
Open WebUI local AI quick start
- 1. Start the model backend and pull a chat modelRun Ollama or another compatible server, load a model, and verify it directly before starting Open WebUI. This separates backend failures from UI configuration.
- 2. Start Open WebUI with a reachable backend URLSet OLLAMA_BASE_URL to the host Ollama address. In Docker use host.docker.internal or the Compose service name instead of localhost inside the container.
- 3. Verify chat, then configure retrieval separatelyConfirm the model appears and completes a chat, then select and test an embedding model for document retrieval. Rebuild stored embeddings if that model or its dimensions change.
Best for
Private chat interfaces over Ollama, local models, or self-hosted model endpoints.
Model support
Open WebUI connects to Ollama through OLLAMA_BASE_URL, or to several servers with OLLAMA_BASE_URLS, and to any OpenAI-compatible server such as LM Studio, llama.cpp or vLLM through OPENAI_API_BASE_URL. Both paths have enable flags and can also be edited in Admin Settings.
Choose models by role
Chat and optional tools
Mistral Small 3.1
Use a chat model that fits the backend and supports any vision or tool features enabled in the workspace. Validate those capabilities per model.
Open the model profile →Document retrieval
Qwen3 Embedding
Use a dedicated embedding model for RAG when the default local embedding path is not the intended choice. Its dimensions must remain consistent with the stored index.
Open the model profile →How to use this integration
- Give a household or team a shared private chat interface over one Ollama server
- Register several model backends and let users switch between them per conversation
- Attach documents for retrieval-augmented chat entirely within your own network
- Manage per-user access to local models through an admin interface rather than shell access
Connecting a local model
OLLAMA_BASE_URL defaults to http://localhost:11434. In Docker that address is the container, so a host-installed Ollama needs http://host.docker.internal:11434, http://host.containers.internal:11434 on Podman, or the service name such as http://ollama:11434 in Compose.
Tradeoffs
One behaviour accounts for most confusion: connection URLs set through Admin Settings are persisted in the database and take precedence over the environment variables, so editing the variable after first boot appears to do nothing. Worth knowing too that ENABLE_OPENAI_API defaults to true with the base URL pointing at OpenAI, which an air-gapped or strictly private install should explicitly turn off.