cloudparts

Open WebUI

A self-hosted web app for chatting with models served by Ollama or any OpenAI-compatible API, with users and groups, document retrieval, tools and web search, under a source-available license that requires keeping its branding above 50 users.

What it is
Running models locally
Used for
Retrieval-augmented generation (RAG)
Type
Commercial product
Made by
Open WebUI, Inc.
License
Open WebUI License since v0.6.6 (April 2025): BSD-3-Clause terms plus a clause that forbids removing or altering the “Open WebUI” branding unless a deployment has 50 or fewer end users in any rolling 30 days, the copyright holder gives written permission, or an enterprise license allows it; the project says it is not OSI-approved. Code up to v0.6.5 stays BSD-3-Clause (MIT before January 2025)
Facts checked
13 October 2026
Website Source code

What it is

Open WebUI is a Python server with a web front end that you run yourself: pip install open-webui then open-webui serve (on Python 3.11), a container image, or Kubernetes with kubectl, Kustomize or Helm. It doesn’t run models itself unless you choose the :ollama image, which bundles Ollama. Instead it connects to Ollama servers and to any OpenAI-compatible endpoint, such as LM Studio, vLLM, OpenRouter, Groq or OpenAI, so one chat window can switch between models or ask several at once. Its data lives in SQLite (optionally encrypted) or PostgreSQL, uploads on local disk, S3, Google Cloud Storage or Azure Blob Storage, and Redis-backed sessions let several workers or nodes run behind a load balancer.

It is multi-user from the start: roles, groups and per-model access, LDAP and Active Directory, OAuth and trusted-header SSO, SCIM 2.0 provisioning, and admin dashboards for messages, tokens and cost. A “model” can be a preset that wraps a base model with a system prompt, tools and knowledge. Knowledge bases do retrieval-augmented generation: an extraction engine (Tika, Docling, Mistral OCR and others) reads each file, the text is embedded into a vector store, and queries use hybrid BM25 and vector search with reranking, or the whole document is injected instead. Tools can be Python code installed in Open WebUI, MCP servers over Streamable HTTP or OpenAPI servers, and filters and pipes can rewrite or route every message.

The license changed with v0.6.6 on 19 April 2025. The code is still public and the terms are BSD-3-Clause plus a branding clause: “Open WebUI” names and logos must stay visible unless the deployment has 50 or fewer end users in any rolling 30 days, the copyright holder gives written permission, or an enterprise license from Open WebUI, Inc. allows rebranding. The project says this means the license would not meet the OSI’s definition of open source. Code up to v0.6.5 remains BSD-3-Clause, and new contributions require signing a contributor license agreement.

Why teams pick it

  • One self-hosted interface for local and hosted models: Ollama plus any OpenAI-compatible endpoint, with side-by-side answers from several models
  • Built for teams: roles and groups, LDAP, OAuth SSO, SCIM 2.0 and usage dashboards, and it runs fully offline when the models are local
  • Retrieval built in: knowledge bases, several extraction engines, hybrid BM25 and vector search with reranking, or whole-document injection
  • Extensible with Python tools and functions, MCP and OpenAPI tool servers, and OpenTelemetry export of traces, metrics and logs

Watch out for

  • Not OSI open source since v0.6.6: deployments above 50 users in 30 days must keep the Open WebUI branding unless they get written permission or buy an enterprise license
  • It is an interface, not an inference engine: speed, model size and cost depend on the Ollama server or API you connect
  • A large attack surface: CVE-2025-64496 let a malicious model server added through Direct Connections (off by default) steal tokens and, chained with the Functions API, run code on the backend until 0.6.35
  • Only ChromaDB and PGVector are maintained by the core team; Qdrant, Milvus, Elasticsearch and the other vector stores are community integrations

Alternatives

What all running models locally have in common →Compare side by side →
AlpacaOpen sourceA GNOME desktop app for Linux that chats with local models through Ollama, which it can install and run for you, and with cloud models through OpenAI-compatible APIs using your own keys.LM StudioCommercialA free, closed-source desktop app and headless daemon that downloads open-weight models from Hugging Face, runs them on your own machine with llama.cpp or Apple’s MLX, and serves them on OpenAI- and Anthropic-compatible endpoints at localhost:1234.

How it fits with other technologies

Run it next to an Ollama server on the same machine · bash

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; the volume keeps the database across upgrades

# or without Docker, on Python 3.11
pip install open-webui
open-webui serve        # http://localhost:8080

See it in a real stack

No architecture uses Open WebUI yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2023–2026

Timothy Jaeryang Baek started the project in October 2023 as Ollama WebUI, a web chat interface for models run by Ollama. On 30 January 2024 Baek announced that the Ollama team had asked for a new name because people took Ollama-webui and its model hub for official Ollama products. The post offered One-webui or Open-webui and a roadmap to use llama.cpp directly and support other runners such as LiteLLM and vLLM; the project became Open WebUI.

It grew from a chat window into a multi-user platform with retrieval, tools and plugins, and a company, Open WebUI, Inc., formed around it. The license went from MIT to BSD-3-Clause in January 2025 and, with v0.6.6 in April 2025, to the Open WebUI License. The team said some groups were stripping the branding, reselling the software and then asking the original developers for support, and that visible branding drives the enterprise licenses and sponsorships that fund the project.

By October 2026 the repository had more than 155,000 GitHub stars and had reached v0.12.0. Companion projects include Open Terminal, which gives models a computer to run code on, and a native desktop app with an optional built-in llama.cpp engine.

Why it’s built this way

Speak to Ollama and to the OpenAI API, not to one vendor
After the rename the project set out to support runners beyond Ollama, and the OpenAI-compatible API became the common path to local servers and cloud providers alike.
Self-hosted, able to run entirely offline
So a person or organization can keep conversations and documents on machines they control, with local models when needed.
A branding clause rather than a closed or SSPL-style license
The team wanted one public codebase for everyone and said it chose this over closing the source, SSPL or BSL terms, or a private enterprise fork.

Timeline

  1. 2023

    Started as Ollama WebUI (October)

  2. 2024

    Renamed Open WebUI after the Ollama team raised trademark confusion (January to February)

  3. 2025

    License changes from MIT to BSD-3-Clause (January)

  4. 2025

    v0.6.6 adopts the Open WebUI License with its branding clause (19 April)

  5. 2025

    CVE-2025-64496 in Direct Connections fixed in v0.6.35 (November)

  6. 2026

    v0.12.0 released (October)

Sources6 · facts checked 13 October 2026

Improve this page on GitHub