Describe your data
Declare the fields, formats, and actions your product actually exposes. That schema drives filters, metrics, grids, charts, validation, and model context.
Open-source building blocks for AI-customizable software
ZUI gives engineering teams the React components, serializable view contracts, and registry-generated LLM prompts to build consoles, internal apps, and reports on their own data—without starting from primitives.
MIT-licensed core Bring your own data and model Optional ZUI Cloud
A real view, not a screenshot
The application owns its data, actions, and allowed components. An engineer, an end user, or an LLM can change the view through the same validated patch model.
How it works
ZUI is the application layer above basic UI primitives: the reusable behavior teams normally spend a quarter rebuilding.
Declare the fields, formats, and actions your product actually exposes. That schema drives filters, metrics, grids, charts, validation, and model context.
Register ZUI components and your own domain components. Engineers and models can compose only the types your application allows.
Write React directly or let an LLM produce a serializable ViewSpec. Both paths render the same real components against the same data.
Human edits and AI suggestions become reviewable patches. Render in your app, save versioned configs, export a PDF, or deliver it on a durable schedule.
Choose an outcome
You can learn the package boundaries later. Each path starts with a complete result.
Connect a row type, then ship filters, metrics, charts, and a production data grid in one working screen.
Open the quickstart →Build a productCompose multiple views with resources, actions, permissions, shared state, approvals, and release controls.
Explore Apps →Deliver a documentUse the same view for a live report and a branded PDF without maintaining a second rendering system.
See PDF export →Build with AI
ZUI publishes focused prompts for common product work. They point the agent at the right packages, constrain it to your real data schema, and require validation before the result is handed back.
Build a ZUI analytics view for the dataset in this project.
Start from the row type and define a DatasetSchema. Register only the ZUI
view types the screen needs. Create a responsive ViewSpec with a filter bar,
headline metrics, one useful chart, and a data grid. Bind every field to the
declared schema; do not invent columns or add raw styling. Validate the final
spec with validateViewSpec and report any data or action adapters I still
need to implement.
One system, not ten libraries
Package boundaries keep the runtime lean. They do not need to become your mental model.
Optional ZUI Cloud modules
Everything needed to build and run the interface stays open source. ZUI Cloud adds Config Store for versioned AppSpec and ViewSpec persistence, PDF Export for production documents, and Scheduled Delivery for recurring Email, SFTP, or Slack handoffs—using one organization and ZUI Cloud API key model.
Your next deliverable