WyrdWerk Deployment Wiki

Open WebUI

3 min read

Ollama gives you the API. Open WebUI gives your team the interface. Without it, only developers can use the model you just spent two weeks configuring.

Open WebUI is a self-hosted, browser-based chat interface for local and cloud LLMs. It runs alongside Ollama or vLLM and gives non-technical team members a ChatGPT-like experience — file uploads, conversation history, model switching — while keeping every query on your hardware.

What It Does

  • Browser UI for any Ollama, vLLM, or OpenAI-compatible backend
  • Connects to multiple providers simultaneously: Ollama, OpenAI, Anthropic, vLLM
  • Built-in RAG: upload documents directly in the chat interface
  • Plugin support, tool calling, and MCP client integration
  • User management, role-based access control, conversation history
  • Embeddable chat widget for internal tools

Installation

docker run -d -p 3000:8080 \
  -v open-webui:/app/backend/data \
  --name open-webui ghcr.io/open-webui/open-webui:main

Open http://localhost:3000. That’s your full chat interface, pointed at whatever Ollama model you have running.

For teams: production deployments use Docker Compose with persistent storage. The official desktop app handles personal installs on Mac and Windows without Docker.

Typical Costs

Open WebUI software is free and open-source. The cost is infrastructure:

SetupCost
Personal laptop (Ollama + Open WebUI)$0
VPS deployment (team of 5-10)~$10–40/month server cost
Dedicated GPU server (RTX 4090)~$1,600 one-time + electricity
Enterprise plan (custom branding, SLA, LTS)Custom pricing

Why It Matters for Compliance Deployments

Open WebUI is the user-facing piece of the self-hosted AI stack. For GDPR or DPDP deployments where queries can’t route through external APIs, the stack becomes:

Team browser → Open WebUI → Ollama/vLLM (your hardware) → Model response

No query leaves your network. No conversation reaches OpenAI’s servers. Open WebUI makes that stack usable for non-engineers.

MCP Support

Open WebUI functions as an MCP client. It connects to MCP servers for tool calls, with support for SSE (Server-Sent Events) transport and auth. This means the same MCP servers you configure for Claude Desktop or Cursor also work in Open WebUI — one server, multiple clients.

Where It Breaks

Concurrency inherited from backend. Open WebUI doesn’t add concurrent request handling — it inherits it from whatever model server it’s pointing at. If Ollama backs it, you’re still capped at ~4 parallel requests. At 10+ simultaneous users, queue times get noticeable. Switch to vLLM for teams above 5 users, or route through LiteLLM to add queuing and load balancing.

Model management is manual. Open WebUI shows you models available in Ollama, but doesn’t manage Ollama itself. Model updates, re-quantization, and version control remain a manual process every 6-8 weeks.

Session auth only. Conversation history is stored locally. No enterprise SSO out of the box on the community build — Enterprise plan required for SAML/LDAP.

When to Choose It

  • You’ve deployed Ollama or vLLM and need team access beyond the API
  • GDPR/DPDP/HIPAA requires queries to stay on-premises
  • You want ChatGPT-like UX for internal tools without SaaS data exposure
  • Teams of 2–20 people where concurrency limits aren’t a bottleneck yet

Don’t use Open WebUI for customer-facing deployments at scale. It’s designed for authenticated internal users, not anonymous production traffic.

Graph View

Backlinks