open-webui/open-webui

Open WebUI – A Self‑Hosted, Extensible AI Front‑End

What it is – Open WebUI is an open‑source, self‑hosted web interface for interacting with LLMs. It works with any OpenAI‑compatible API (OpenAI, Azure, Groq, Mistral, etc.) and with local models served by Ollama, letting you run AI completely offline or mix cloud and on‑premise providers.

Why it matters – Most chat‑style AI tools are SaaS‑only, lock you into a single provider, and require an internet connection. Open WebUI gives you a full‑featured, browser‑based UI that you can run on a laptop, a server, or in Kubernetes, with fine‑grained access control, plugins, RAG, multi‑model chats, and even a built‑in terminal for agents to execute code.

Core Features (as listed in the README)

Easy installation – pip install open-webui, Docker images (:ollama, :cuda), Docker‑Compose, Helm/Kustomize for Kubernetes.
Provider‑agnostic model access – Connect any OpenAI‑compatible endpoint and local Ollama models; switch between LMStudio, GroqCloud, Mistral, vLLM, etc.
Granular RBAC & groups – Admins define roles, permissions, and per‑group experiences.
Plugin system – Filters, Actions, Pipes, Tools, Skills; extend via MCP/MCPO or OpenAPI servers.
Agents & “Open Terminal” – Give a model a sandboxed terminal/filesystem to run scripts, fix errors, and produce files. Enterprise version adds per‑user isolated containers.
Notes & Channels – Persistent note workspace, shared real‑time channels with threads, reactions, pins.
Persistent memory – The AI can retain facts about a user across chats.
Live workflow UI – Checklists, queued messages, real‑time progress.
Calendar & scheduling – Built‑in calendar with function‑calling for AI‑driven scheduling.
Automations – Schedule prompts, view runs on the calendar, link back to chat output.
Responsive PWA – Works on desktop, laptop, and mobile; offline on localhost.
Full Markdown & LaTeX – Rich text rendering.
Voice/Video calls – Integrated STT (Whisper, OpenAI, Deepgram, Azure) and TTS (Azure, ElevenLabs, OpenAI, Transformers).
Artifact storage – Key‑value API for persisting files, leaderboards, trackers.
Local RAG – 9 vector DB options, 10+ content extractors, hybrid BM25+vector search, # command to pull docs.
Web search for RAG – Plug‑in dozens of search providers (SearXNG, Google PSE, Kagi, Bing, etc.) and inject results.
Web browsing – Pull a URL into chat via # or let the model fetch it.
Image generation & editing – DALL·E, Gemini, ComfyUI, AUTOMATIC1111 (local).
Multi‑model conversations – Run several models side‑by‑side.
Analytics & evaluation – Admin dashboards, token/cost tracking, arena, A/B testing, ELO leaderboards.
Flexible persistence – SQLite (optional encryption) or PostgreSQL; file storage locally or on S3/Google Cloud/Azure.
Advanced vector DBs – Chroma, PGVector, Qdrant, Milvus, Elasticsearch, OpenSearch, Pinecone, S3Vector, Oracle 23ai.
Enterprise auth – LDAP/AD, SSO via headers/OAuth, SCIM 2.0 provisioning (Okta, Azure AD, Google Workspace).
Cloud‑native file pickers – Google Drive, OneDrive/SharePoint.
Observability – OpenTelemetry traces, metrics, logs.
Horizontal scalability – Redis‑backed sessions, WebSocket support for multi‑node deployments.
i18n – Multilingual UI.
Security process – Public security policy and responsible‑disclosure.

Ecosystem (companion projects)

Project	What it adds
Open WebUI Computer (`open-webui/computer`)	Mobile‑first coding/terminal agent that can be used as a model or via messaging apps.
Open Terminal / Terminals (Enterprise) (`open-webui/open-terminal`, `open-webui/terminals`)	Self‑hosted containerized terminals that the AI can drive; enterprise version gives per‑user isolation and resource limits.
oikb (`open-webui/oikb`)	Syncs knowledge bases from 45+ sources (GitHub, Confluence, ServiceNow, Salesforce, Jira, Slack, Notion, etc.) into the RAG pipeline.
Desktop App (`open-webui/desktop`)	Native macOS/Windows/Linux client with system‑wide Spotlight chat bar, screenshot capture, push‑to‑talk, and optional local llama.cpp inference.

Getting Started (quick‑start commands from the README)

# Python install (requires Python 3.11)
pip install open-webui
open-webui serve   # → http://localhost:8080

# Docker (CPU only, Ollama on host)
docker run -d -p 3000:8080 \
  -v open-webui:/app/backend/data \
  --add-host=host.docker.internal:host-gateway \
  --name open-webui --restart always \
  ghcr.io/open-webui/open-webui:main

# Docker with bundled Ollama (CPU)
docker run -d -p 3000:8080 \
  -v ollama:/root/.ollama -v open-webui:/app/backend/data \
  --name open-webui --restart always \
  ghcr.io/open-webui/open-webui:ollama

# Docker with GPU (CUDA image)
docker run -d -p 3000:8080 --gpus all \
  -v ollama:/root/.ollama -v open-webui:/app/backend/data \
  --name open-webui --restart always \
  ghcr.io/open-webui/open-webui:cuda

The README also mentions Docker‑Compose, Helm, Kustomize, and a --network=host flag for troubleshooting connection issues.

Typical Use Cases

Private LLM playground for teams that need to keep data on‑premise.
Enterprise AI assistant with calendar, scheduling, and document‑aware chat.
RAG‑powered knowledge base that pulls from internal docs, vector stores, and live web search.
AI‑driven coding via the Open Terminal or the Computer agent.
Multi‑model evaluation using the built‑in arena and analytics dashboards.
Custom extensions through plugins, OpenAPI tools, or community‑built skills.

License & Governance

The repository contains multiple licenses. Core code is under the Open WebUI License (which requires preserving the “Open WebUI” branding) plus the original licenses of contributed components. Detailed terms are in LICENSE and LICENSE_HISTORY.

Community & Support

Discord: https://discord.gg/5rJgQTnV4s
Documentation: https://docs.openwebui.com/
Issue tracker on GitHub for bugs & feature requests.
Security: Responsible‑disclosure via GitHub security tab.
Sponsorship: GitHub Sponsors link in the README.

TL;DR

Open WebUI is a full‑featured, self‑hosted web UI for LLMs that supports both local (Ollama) and cloud APIs, offers enterprise‑grade auth, RBAC, plugins, RAG, a terminal for agentic execution, and a growing ecosystem of companion tools. It can be run with a single pip command or via Docker/Kubernetes, making it a practical choice for anyone who wants control over their AI stack.