OpenUI: Text to Live UI Preview

A button, a form or a settings panel rarely takes long because of its logic. The slow part is adjusting spacing and rewriting HTML and CSS again and again. OpenUI, from machine learning platform Weights & Biases (W&B), lets you describe the interface in text, renders it immediately, and keeps refining it through plain-language requests such as “add a dark mode.” You can then export the result as React, Svelte or Web Components code. W&B uses it internally to prototype its own tooling for LLM applications. The repository is Apache-2.0 licensed and had 22,568 stars as of 2026-10-10.

Core features

• Text in, live preview out: Describe an interface and it renders at once. If something looks wrong, say what to change instead of hunting through markup.
• Export to several frameworks: The README says HTML can be converted to React, Svelte, Web Components and similar targets, which cuts the rewrite between prototype and project.
• Choice of model providers: OpenAI, Groq, Gemini, Anthropic, Cohere and Mistral work out of the box. LiteLLM extends support to more providers, and OpenAI-compatible endpoints such as LocalAI also work.
• Local models through Ollama: The README’s example uses llava, one of the few Ollama models that accept images, so you can keep both the model and the prompts on your own machine.
• One-command Docker start: The official image is ghcr.io/wandb/openui, listening on port 7878. Pass your API keys as environment variables and you are running.
• A ready development setup: The repo includes a dev container, plus Codespaces and Gitpod configurations for hot-reloading both frontend and backend.

Typical use cases

• Full-stack developers sketch an admin dashboard in minutes, judge the layout, then decide which parts are worth writing by hand.
• Product managers and designers turn an idea into a clickable demo and discuss the interaction in a review, instead of arguing over a text description.
• Teams with strict privacy rules run a local model through Ollama inside their network, so requirements and generated code stay in-house. The README warns that local runs without a GPU will be very slow.

Quick start

The README’s recommended Docker command is below. Replace the xxx placeholders with your own keys:

export ANTHROPIC_API_KEY=xxx
export OPENAI_API_KEY=xxx
docker run --rm --name openui -p 7878:7878 -e OPENAI_API_KEY -e ANTHROPIC_API_KEY -e OLLAMA_HOST=http://host.docker.internal:11434 ghcr.io/wandb/openui

Open http://localhost:7878 and describe your first interface. To try it without installing anything, the hosted demo at openui.fly.dev is linked from the README.

Summary

Good for developers and product teams who want to test interface ideas quickly and keep control of the code. For judgment on visual style, pair it with the frontend-design skill. The project is Apache-2.0 licensed, written mainly in TypeScript, and was last pushed on 2026-10-09.

A few points to weigh first:

• The README admits it is “not as polished” as v0. Treat its output as a prototype, not production UI.
• Hosted models need your own API keys and are billed by usage. Running fully offline means supplying a capable local model, and that takes serious hardware.
• Generated code still needs review, especially for interaction logic and data handling. The README makes no promises about either.