Easiest AI Chat UIs to self-host

Last reviewed Oct 11, 2026 · 634 words

Open WebUI is the easiest AI chat interface to self-host and the right first install for almost everyone. It is the only app in this category with a verified single-container compose file, the directory rates it Easy, and it grows from one user to a household without a migration. The catch a beginner rarely sees coming is that the 2 GB estimate covers the interface only. The model you chat with is a separate program, and that is where the real hardware bill lives.

Why not the lighter picks?

NextChat and ChatBot UI rank next because they ship one-click platform packages, but neither suits a first self-hosted install. NextChat is single-user focused and expects provider API keys; it is a fine personal client, not a household service. ChatBot UI is listed in the catalogue as no longer actively maintained, and in a category where provider APIs change often, a frozen frontend breaks quietly. Skip it.

The genuinely lighter option is Hollama, estimated at 256 MB with no database: conversations live in your browser's storage. That is its appeal and its limit. Nothing on the server to back up, but nothing synced between your laptop and phone either, and clearing the browser clears your history. The Hollama vs Open WebUI comparison covers the trade.

What "easy" still leaves you to do

First, a backend. Open WebUI talks to Ollama on your own hardware or to any OpenAI-compatible API, and you install and pay for that part separately. Second, persistence: the catalogue's deploy config mounts /app/backend/data, which holds accounts, chat history and uploaded documents, so put it on a named volume and include it in your backups. Third, accounts: the first account created becomes the admin, so claim it before the URL leaves your machine. Fourth, the reverse proxy. If the UI is reachable from outside your network, it needs HTTPS and a login in front; an exposed chat UI wired to a paid API key lets strangers spend your credits.

How much RAM does a self-hosted ChatGPT really need?

Plan two budgets. The interface: 2 GB for Open WebUI by the directory's estimate, far less for the lighter clients. The model: if you run it locally, its size sets the requirement, and even small models want several gigabytes of RAM or VRAM on top of the UI. The VRAM math post works through the numbers by model size. If you use a hosted API instead, the model costs you no memory at all, and a modest VPS runs the whole thing.

When to step up

Move to LibreChat (Medium, 2 GB) when a team arrives and wants OAuth or LDAP logins, agents, and switching between many providers; it needs MongoDB alongside it, which is the extra work you are signing up for. Move to RAGApp only if document question-answering is the whole goal and you want a no-code setup for it. Most households never outgrow Open WebUI.

What I'd do

Open WebUI in one container with its data directory on a named volume, Ollama beside it, your own account created first, and HTTPS in front before anyone else gets the link. Pick the model to fit the RAM you have left after that, not the other way round.