Open WebUI
Self-hosted AI interface that connects to any model, local or cloud
About Open WebUI
Open WebUI is the interface a lot of teams put in front of their own inference stack. It runs on your hardware, connects to Ollama, OpenAI, Anthropic or any OpenAI-compatible endpoint, and gives you multi-user chat, retrieval over your own documents, image generation, voice, and a plugin system in Python. The reason it exists is straightforward: plenty of organisations want the ChatGPT experience but cannot send their prompts to a vendor, and the alternative is a patchwork of separate tools.
One thing has to be said clearly, because the star count predates it. Open WebUI is no longer open source in the OSI sense. From version 0.6.6, released in April 2025, the BSD-3 licence carries an added branding-protection clause: you may not remove or alter Open WebUI branding unless your deployment has 50 or fewer users in a 30-day window, you are a substantive contributor with written permission, or you hold an enterprise licence. Everything up to and including v0.6.5 remains BSD-3 and can be forked with no restriction, which the project documents openly and points to as the escape hatch. For a small team or an internal deployment under 50 users, nothing changes. For a company planning to white-label it, it does, and enterprise pricing is not published anywhere on their site.
Deploy with Docker, Kubernetes or a Python install, then add model connections: a local Ollama instance, an API key for a hosted provider, or any OpenAI-compatible URL. Users sign in against your identity provider and are placed into groups that control which models and tools they can reach. Conversations support switching model mid-thread and running two models side by side to compare. Documents uploaded into a knowledge base are chunked and indexed into one of the supported vector databases, or injected whole in full-context mode when precision matters more than token cost. Capability is added rather than configured: Python tools run inside the chat with a built-in editor, pipelines filter or route every message, and MCP and OpenAPI servers are discovered and exposed as tools automatically. The community site hosts prompts, tools and model presets that install with one click.
- •Any Model, One Interface - Ollama, OpenAI, Anthropic and any OpenAI-compatible provider side by side, with mid-conversation switching and two-model comparison
- •Knowledge and RAG - Hybrid BM25 and vector search with cross-encoder reranking, 13 vector database integrations, 8 document extraction engines, and a full-context mode that skips chunking
- •Python Extensibility - Tools and functions written in Python and edited in the browser, plus pipelines, native MCP over Streamable HTTP, and auto-discovery of OpenAPI servers
- •Model Presets as Agents - Wrap any base model with a system prompt, bound tools, knowledge and per-group access control to produce a specialised assistant
- •Open Terminal - A real execution environment attached to the chat, where the model runs commands, reads output, fixes errors and iterates, in a container or on bare metal
- •Enterprise Deployment - Multi-user from the first install, with SSO, role-based access control, audit logs, data residency and air-gapped on-premises deployment
Organisations that need AI available to staff without prompts leaving their network, and developers who want one front end over a mixture of local and hosted models. It is a strong fit for regulated industries and for teams already running Ollama or vLLM who need an interface, access control and document search on top. Their own documentation names Samsung Semiconductor, Astellas Pharma with more than 3,000 employees, and Public Storage as deployments. It is a weaker fit in two cases. If you want to rebrand it for more than 50 users, the licence blocks that without an enterprise agreement whose price is not published, and openwebui.com/enterprise returned a 404 upgrade notice when this entry was written, so the only route is a contact form. And if you want a coding agent rather than a chat platform, this is not that: it is an interface layer, and the code work happens in tools like Codex, Cline or OpenCode.
Pricing
Custom pricing, contact sales
- Standard Use (self-hosted, free)$0/mo
- Enterprise License (contact sales)Contact sales
From the vendor pricing page, 2026-09-15
