open-webui/open-webui

What it solves

Open WebUI solves the challenge of managing and interfacing with multiple AI models from different providers in one place. Without it, users often need separate tools for local models (like Ollama) and cloud APIs, and they lack a unified, feature-rich chat experience. The project also addresses privacy and control needs by allowing full self-hosting and offline operation.

How it works

Open WebUI is a self-hosted web application that you install on your own server or computer. It connects to model backends through two main routes: local models via Ollama, and any OpenAI-compatible API (such as LMStudio, GroqCloud, OpenRouter, vLLM, and more). The platform provides a chat-like interface where you can interact with one or multiple models at once. It includes a plugin system (Filters, Actions, Pipes, Tools, Skills) for custom integrations, a built-in RAG system with support for 9 vector databases and web search, persistent memory across conversations, and agentic execution via a terminal environment. It also offers role-based access control, usage analytics, and deployment options via pip, Docker, or Kubernetes.

Who it's for

This is for individuals and organizations that want a powerful, self-hosted AI assistant interface. It suits power users who want to mix local and cloud models, teams that need fine-grained access control and enterprise authentication (LDAP, SSO), and developers who want to extend the platform with plugins. It's also for those who prioritize data privacy and want to run AI fully offline.

Highlights

• Easy installation via pip, Docker, kubectl, or Helm, with images for Ollama and CUDA.
• Broad model support: connect any OpenAI-compatible API alongside local Ollama models.
• Granular role-based access control (RBAC) and user groups.
• Extensible plugin system with Filters, Actions, Pipes, Tools, and Skills, plus MCP/OpenAPI support.
• Build specialized agents with custom instructions, tools, and knowledge.
• Agentic execution with a built-in terminal (Open Terminal) for multi-step tasks.
• Persistent memory that carries user facts across conversations.
• Local RAG with 9 vector databases, hybrid search, reranking, and web search injection.
• Web browsing with # command to fetch URLs.
• Image generation and editing via multiple engines (DALL·E, Gemini, ComfyUI, AUTOMATIC1111).
• Multi-model conversations in parallel.
• Usage analytics and model evaluation with arena, A/B testing, and ELO leaderboards.
• Flexible database choices (SQLite or PostgreSQL) and file storage (local, S3, GCS, Azure Blob).
• Enterprise features: LDAP, SSO, SCIM 2.0 provisioning, OpenTelemetry observability, horizontal scaling.
• Responsive design and PWA, voice/video calls with multiple STT/TTS providers.
• Full Markdown and LaTeX support.