AI Playground by hjLabs.in

An Open WebUI alternative with no server behind it

Open WebUI is the self-hosted front end most people put in front of Ollama: a polished multi-user chat interface with document RAG, model management and access control, usually run as a Docker container next to a backend that actually holds the weights. It is excellent, and it presumes you are willing to operate two services. This page is the opposite trade: a chat interface with no backend at all, because the model is downloaded into your browser and executed on your GPU. Nothing to deploy, nothing to keep patched, and no container that has to stay up for the URL to work.

At a glance

Deployment
None — open a URL
Backend required
No; Open WebUI needs one (Ollama, vLLM, an OpenAI-compatible API)
Where the model runs
The visitor’s own GPU, in the tab
Multi-user accounts
No — Open WebUI has them
Document RAG
No — Open WebUI has it
Chat history
Local to the browser, not synced
Model ceiling
Roughly 0.5B–3B parameters

Where Open WebUI is clearly better: it is a system, not a page. Multiple accounts with permissions, conversation history that survives on a server you own, document upload and retrieval, tool calling, connections to any OpenAI-compatible endpoint, and — through that backend — models far larger than a browser can hold. For a team standardising on self-hosted AI, that is the correct choice and this is not.

Where no-server wins: there is no server. No Docker, no compose file, no reverse proxy, no TLS to renew, no port exposed on a home network, and no update treadmill on a service holding your conversations. The security surface of a page that runs a model in a sandboxed tab is very different from that of a long-lived container with an admin login.

The honest limits are the mirror image of that. Conversations live in your browser, so they do not follow you between devices and there is no shared history for a team. There is no document library and no retrieval pipeline. And the model ceiling is roughly 0.5B to 3B parameters rather than whatever your server can hold. If any of those three is a requirement, run Open WebUI.

open webui alternativechat ui without dockerself hosted llm ui alternativellm web interface no serverollama web ui alternativebrowser llm interfaceno backend chat uiprivate llm chat no install

Frequently asked questions

Do I need Ollama or any backend running?

No. There is nothing for this page to connect to — the weights are fetched from Hugging Face directly by your browser and executed locally. That is also why it works on a machine where you cannot install or run a daemon.

Can several people share it?

They can each open it, but they do not share anything: no accounts, no shared history, no shared model store. Every visitor downloads the model into their own browser and runs their own copy. If you need shared state, you need a server, which means Open WebUI.

Can I upload documents and ask questions about them?

Not as a retrieval pipeline. You can generate embeddings locally on the embeddings page and do question answering over a passage you paste in, but there is no document library, no chunking and no vector store — that is a genuine Open WebUI feature this does not have.

Where is my chat history kept?

In your browser, on your device. It is not synced to a server, which means it is private by construction and also that it does not follow you to another machine.

Is my data uploaded to a server?

No. The model weights are downloaded from Hugging Face to your browser once, and every inference after that runs on your own GPU through WebGPU. Your text, images and audio are never sent anywhere — hjLabs.in has no inference server and no way to see your input.