Open WebUI

The dominant ChatGPT-clone front-end for local LLMs — multi-user, RAG, MCP, plugins, the canonical pair with Ollama.

★ ★ Open WebUI (github.com/open-webui/open-webui, formerly Ollama-WebUI) is the dominant FOSS chat front-end for self-hosted AI in May 2026. The canonical pair with Ollama. Multi-user with auth, built-in RAG, MCP support, image gen integration, plugin system, ChatGPT-grade UX, AGPL-licensed.

If you've read the overview and the hardware tier guide, this is the chat UI you almost certainly want. For desktop apps see Jan / GPT4All or LM Studio; for multi-provider routing see LibreChat / AnythingLLM.

What it is

Web app, deployed via Docker / pip install open-webui / Helm.
OpenAI-compatible client — talks to Ollama, vLLM, LM Studio, OpenAI, Anthropic, OpenRouter, Groq, Mistral, anything OpenAI-compat.
Multi-user with auth — sign-up flow, admin panel, role-based permissions, group-level model access.
Built-in RAG — drop documents into a knowledge base, chat over them.
Web search integration — SearXNG, Brave, Tavily, Serper, Google PSE — model gets to "search the web."
Image generation — Automatic1111 / ComfyUI hooks.
Voice in/out — STT (Whisper), TTS (Piper / OpenAI TTS / Kokoro).
Pipelines / Functions — Python plugin system for adding tools, custom RAG, filters.
MCP (Model Context Protocol) support — added 2024–25.
License: AGPL-3.0 with a (relaxed in 2025) clause for non-modifying redistribution.

Quick start (Docker)

docker run -d -p 3000:8080 \
  -v open-webui:/app/backend/data \
  --add-host=host.docker.internal:host-gateway \
  -e OLLAMA_BASE_URL=http://host.docker.internal:11434 \
  --restart always \
  --name open-webui \
  ghcr.io/open-webui/open-webui:main

Visit http://localhost:3000, create an admin account, point at your Ollama instance, start chatting.

Major features in May 2026

★ Workspaces — separate per-team / per-project chat environments.
★ Knowledge bases — multiple per workspace; vector store (Chroma / Qdrant / pgvector) under the hood.
★ Tool integration via Pipelines and MCP servers.
★ Channels — real-time multi-user chat rooms with the model as a participant; useful for team brainstorming.
★ Markdown / LaTeX / Mermaid / code-rendering — full ChatGPT-equivalent rendering.
★ Code interpreter — sandboxed Python execution for the model.
★ Document upload + chat — PDF, DOCX, image-with-OCR.
★ Custom models — chain together a model + system prompt + knowledge + tools = a named "custom model" reusable across chats.
Model evaluation arena — A/B compare two model responses side-by-side.

RAG

Open WebUI ships with a serviceable RAG pipeline:

Embeddings model (default all-MiniLM-L6-v2; configurable).
Vector store (Chroma default; pgvector / Qdrant / Milvus supported).
Hybrid keyword + vector search.
Reranker support — see embeddings & rerankers.
File upload UI; URL scraping; YouTube transcript ingestion.

For RAG-first workflows or larger document corpora, see ai-selfhost-rag-local — Open WebUI's RAG is good enough for most users; AnythingLLM, LightRAG, Khoj are more RAG-specialized.

Honest limits

AGPL caveats. If you embed Open WebUI in a commercial product, read the license carefully. For personal / internal-team use, fine.
Heavyweight container. ~1GB image; runs Postgres / SQLite; not a "tiny" install.
Plugin sprawl risk. Like Obsidian, you can over-install Pipelines and end up with a fragile setup.
Mobile-first it isn't. Works on mobile but designed for desktop.
Update cadence is fast — weekly releases sometimes; pin a version in production.

Common patterns

Open WebUI + Ollama on a single home server / VM. The canonical setup.
Open WebUI + vLLM — point Open WebUI at vLLM's OpenAI endpoint for production multi-tenant.
Open WebUI behind Authelia / Authentik — SSO for the family / team.
Open WebUI + LiteLLM — proxy to multiple model providers with budgets and logs.
Open WebUI + ComfyUI — chat that can also generate images mid-conversation.

Pick this if…

The default chat UI for self-host: ★ ★ Open WebUI.
You want multi-user with auth: Open WebUI.
You want RAG built-in: Open WebUI (or AnythingLLM if RAG is the main focus).
You're a single developer wanting a desktop app: Jan or LM Studio.
You need multi-provider with charging / budgets: LibreChat.