GitHub – huggingface/chat-ui: Open source codebase powering the HuggingChat app

Chat UI from Hugging Face: a practical chat interface for OpenAI-compatible LLMs

You walk into a codebase and see a clear path to a running chat app.

You wonder how much work it will take to connect your models.

This article reviews the Hugging Face Chat UI repository and what it means for teams exploring GenAI chat experiences.

Why this matters to you.

– It lowers the friction to run a chat front end against any OpenAI-compatible model.

– It maps directly to typical production needs like MongoDB storage and container deployment.

– It helps you assess integration effort before committing engineering time.

What Chat UI is and how it works

Chat UI is a SvelteKit application that powers the HuggingChat experience on hf.co/chat.

It intentionally speaks only to OpenAI-compatible APIs via OPENAI_BASE_URL and the /models endpoint.

Why that choice?

Because any service that implements the OpenAI protocol — for example llama.cpp servers, Ollama, or OpenRouter — will work by default.

What is removed from this build?

Provider-specific helpers like legacy MODELS env usage, GGUF discovery, and specialized embedding or web-search plugins.

Example prompt snippet to try in your environment.

OPENAI_BASE_URL=https://api.your-openai-compatible-endpoint.com
OPENAI_API_KEY=your_api_key_here

I recently cloned the repo and swapped OPENAI_BASE_URL to a local Ollama endpoint.

It responded immediately in the UI.

Deployment and integration notes you should know

The repo expects a MongoDB backend for chat history, users, settings, files, and stats.

You can run MongoDB locally or use a managed cluster such as MongoDB Atlas.

Which env vars matter most?

– MONGODB_URL points to your MongoDB URI.

– MONGODB_DB_NAME is optional for custom DB naming.

– OPENAI_BASE_URL and OPENAI_API_KEY configure the model provider.

How to get started quickly?

– Create a .env.local from the template in the root.

– Populate OPENAI_BASE_URL and OPENAI_API_KEY for an OpenAI-compatible service.

– Set MONGODB_URL to a hosted Atlas URI or mongodb://localhost:27017 for a local container.

– Run the dev server with npm run dev and preview production with npm run preview.

Prefer containers?

You can run everything in one container as long as you supply a MongoDB URI via -e flags.

A practical reference and the source code live on GitHub.

You can explore the repo here: https://github.com/huggingface/chat-ui

Final thoughts.

Chat UI is a pragmatic, lightweight front end for teams that want a familiar chat experience tied to OpenAI-compatible back ends.

It removes many early decisions while letting you choose your model router and database.

What does this mean for business value?

It shortens the path from prototype to production for conversational features.

It also lets product and engineering teams test model choices without rewriting the UI layer.

Will you try it in a sandbox first?

I recommend running it against a routed Hugging Face Inference Providers endpoint or a local Ollama/llama.cpp setup to validate latency and cost.

Generative AI is moving from research demos to integrated product features.

This repo is one practical building block for that transition.

Full repo: https://github.com/huggingface/chat-ui