Open WebUI
Run a self-hosted chat UI with Ollama, OpenAI-compatible APIs, local RAG, and optional offline operation.
What it is
Open WebUI is an AI chat platform you can deploy yourself. Users open a browser, pick a connected Ollama, local model, or OpenAI-compatible API, and after sending a question the platform passes the request to the corresponding model and saves the conversation; when documents need to be handled, document or web page content can also be retrieved and brought into the current conversation.
What sets it apart from using the Ollama interface, different vendors' web pages, or writing a separate API client is that it brings models, users, permissions, memory, knowledge bases, and extensions together in one interface. Administrators can control who has access to which models, knowledge bases, and tools, while users can use local and remote models within a single conversation.
How to run
Install and start it in a Python 3.11 environment:
pip install open-webui
open-webui serveAfter starting, visit http://localhost:8080. You can also run the official Docker image; when deploying, /app/backend/data must be mounted, otherwise the database may be lost when the container is rebuilt.
Core mechanisms
- Unified model entry point: connects to Ollama and OpenAI-compatible interfaces at the same time, mixing local and remote models in one interface.
- Built-in RAG: through a vector database, content extractors, hybrid retrieval, and reranking, it finds relevant content from documents or web pages and adds it to the conversation.
- Extension system: Filters, Actions, Pipes, Tools, and Skills can adjust the message flow or connect external services, and can also connect to MCP and OpenAPI tool servers.
- Multi-user governance: uses roles, user groups, and fine-grained permissions to limit access to models, knowledge bases, and extensions.
Who it's for
It suits people who want to build a unified LLM chat entry point for themselves or their team, and administrators who need to centrally manage local Ollama, enterprise APIs, document libraries, and identity systems. Simple deployments can use SQLite; at scale, it can work with PostgreSQL, object storage, Redis, and multi-node capabilities.
Usage boundaries
Before running offline, you need to prepare the models, embedding models, and related resources locally and set HF_HUB_OFFLINE=1. Network access is still needed when connecting to cloud APIs, search services, or external storage.
The current code includes multiple licenses. The Open WebUI License requires keeping the "Open WebUI" branding; capabilities such as custom branding, SLA, and LTS are part of the Enterprise Plan. Before deploying or modifying, check LICENSE and LICENSE_HISTORY.