Every model in one interface
OpenAI, Anthropic and Gemini, switched in one screen.
Self-host Open WebUI · Agent Lab Deploy
Self-host Open WebUI.
Your team's own ChatGPT.
An open-source ChatGPT: say one sentence to your AI tool, Agent Lab runs Open WebUI on PostgreSQL, add your model key, and the team is in.
> Set up Open WebUI from agentlab.cresclab.com/en/deploy/open-webui
sites_createchat → chat--acme.hosting.agentlab.run · medium
databases_createPostgreSQL 18 · S · pgvector
databases_connect_appDATABASE_URL, PGVECTOR_DB_URL set
sites_secrets_generateWEBUI_SECRET_KEY · generated
Open WebUI is up: chat--acme.hosting.agentlab.run. The first person to sign up is the admin; invite the team from the admin panel.
Why self-host
ChatGPT Team bills per seat. Self-hosted bills per use.
Knowledge in PostgreSQL
Vectors in pgvector, in the same database as accounts and chats.
Team accounts you control
Public sign-up off; admins invite colleagues and set model access.
Upgrades roll back
If an upgrade breaks, one action restores the previous version.
Choosing
ChatGPT Team vs a VPS vs Agent Lab
The table scrolls sideways.
| What matters to you | Agent Lab + Open WebUI | ChatGPT Team | Your own VPS |
|---|---|---|---|
| Billing | Hosting plus actual model usage | Fixed per seat per month | Hosting plus model usage |
| Models | OpenAI, Claude, Gemini | OpenAI | Any |
| Where the data lives | Your Agent Lab organization | OpenAI | Your server |
| Who runs it | The platform | OpenAI | You |
| Setup | One sentence to an AI tool | Sign up | Docker, reverse proxy, TLS |
How
Connect → one sentence → model key → invite the team
Needs size M or larger, so it starts on Pro.
Connect Agent Lab
One MCP line in your AI tool, one browser sign-in.
Say one sentence
The agent creates the app, connects PostgreSQL with pgvector and generates WEBUI_SECRET_KEY.
Add a model key
Enter OPENAI_API_KEY as a console project secret.
Invite the team
The first sign-up is the admin; invite colleagues from the admin panel.
Copy and go
The Dockerfile and the sentence
Everything lives in PostgreSQL and embeddings go through an API, so a rebuild loses nothing.
Dockerfile
FROM ghcr.io/open-webui/open-webui:main
ENV DATA_DIR=/tmp/open-webui \
STATIC_DIR=/tmp/open-webui/static \
VECTOR_DB=pgvector \
RAG_EMBEDDING_ENGINE=openai \
RAG_EMBEDDING_MODEL=text-embedding-3-small \
ENABLE_SIGNUP=false \
SCARF_NO_ANALYTICS=true \
DO_NOT_TRACK=true
EXPOSE 8080
# Open WebUI reads $PORT; DATABASE_URL and PGVECTOR_DB_URL come from the platformDATABASE_URLwritten by databases_connect_appPGVECTOR_DB_URLsame database, holds vectors; needs pgvectorWEBUI_SECRET_KEYgenerated by sites_secrets_generate; changing it signs everyone outOPENAI_API_KEYyour model key, a console project secret
Say this to your AI tool
Read https://app.agentlab.cresclab.com/agent/deploy.md and deploy Open WebUI to Agent Lab with the Dockerfile at agentlab.cresclab.com/en/deploy/open-webui: create and connect a PostgreSQL with pgvector, point DATABASE_URL and PGVECTOR_DB_URL at it, generate WEBUI_SECRET_KEY on the platform, and tell me where in the console to enter OPENAI_API_KEY. Use size M. Give me the URL when it's up.FAQ
Before you self-host Open WebUI
How much does it cost to self-host Open WebUI?
It starts on Pro: US$29 a month before tax with US$14 of monthly usage credit, which the app draws on. Model usage is paid to OpenAI, Anthropic or Google as you use it.
Can Open WebUI use Claude and Gemini?
Yes. Add Anthropic and Gemini as admin-panel connections or through an OpenAI-compatible gateway, and switch in one interface.
Do I need Ollama for Open WebUI?
No. Ollama needs a lot of memory or a GPU; connect OpenAI, Anthropic or Gemini APIs directly and medium is enough.
Where does Open WebUI store its data?
Accounts, chats, settings and vectors are in PostgreSQL with pgvector. The container only writes to /tmp, which clears on rebuild.
Can my whole team use it?
Yes. The first sign-up is the admin; set ENABLE_SIGNUP to false, invite colleagues and set which groups use which models.