A self-hosted AI platform that refuses to stay in its lane

What it does

Open WebUI is a browser-based control panel for chatting with local and remote language models. It plugs into Ollama and any OpenAI-compatible API, then layers on RAG document search, image generation, voice calls, and role-based access control. The project aims to be a complete, offline-capable AI workstation you host yourself.

The interesting bit

The feature list reads like someone asked “what if we bolted an entire SaaS platform onto a chat window?” and actually shipped it. Native Python function calling, a built-in model builder, SCIM 2.0 provisioning, and OpenTelemetry observability all live in the same UI that also runs ComfyUI image generation and hands-free video calls.

Key highlights

• Runs fully offline with Ollama or online with OpenAI-compatible endpoints such as LMStudio, Groq, and OpenRouter
• Local RAG with support for 9 vector databases and 15+ web-search providers, plus document extraction via Tika, Docling, and Mistral OCR
• Enterprise auth including LDAP/Active Directory, OAuth, and SCIM 2.0 automated user provisioning
• Built-in image generation and editing through DALL-E, Gemini, ComfyUI, and AUTOMATIC1111
• Horizontal scaling via Redis, with PostgreSQL or SQLite backends and S3-compatible cloud storage

Caveats

• Python 3.11 is strictly required for pip installs; other versions are explicitly flagged as incompatible
• Docker deployments must mount a specific data volume to avoid database loss—a requirement the README labels as “crucial”
• The PWA offers offline access only on localhost, so remote mobile usage still needs connectivity

Verdict

Ideal for teams that need a private, multi-user AI workspace with LDAP, RAG, and image generation under one roof; avoid it if you just wanted a thin wrapper around a single model endpoint.