Alpha. Not on crates.io yet.

The AI-native Rust framework for data-heavy business apps

AI agents write the code. People review it. The framework keeps it safe.

What AI-native means here

Every action is a typed tool

Each #[action] is an API endpoint and a tool for AI agents. Its JSON schema comes from the input type, its description from the doc comment.

Models stay inside your permissions

Agents act as the signed in user. ai = false hides an action from models, confirm makes the user approve it first, and every agent call is audited.

Compiler errors written for agents

A missing permission, a missing translation or a wrong field type fails at compile time, with a message that says what to change.

An MCP server for your project

zeyn mcp gives coding agents the docs, a structural check of the app and its tests as tools.

Built on axum, not instead of it

axum handles HTTP. Zeyn adds what a business app needs on top of it, the way Django sits on WSGI.

You get	axum	Zeyn
Routing, middleware, async HTTP	Yes	Yes (it is axum)
Models, validation on deserialize, row versions	No	Yes
Server rendered screens, DataGrid, forms, pickers	No	Yes
Permissions on every action, CSRF, strict CSP, audit, masking	No	Yes
Translations checked at compile time, time zones, money	No	Yes
API with OpenAPI, MCP tools, embedded assistant	No	Yes

A list, an editor and an API

#[action(permission = "sales.order.read")]
pub async fn search_orders(q: DataQuery, store: OrderStore) -> Result<Page<Order>> {
    Ok(q.apply(store.all()?))
}

#[screen(route = "/", permission = "sales.order.read")]
pub fn order_list() -> Screen {
    Screen::new(t!("orders.title")).body(
        DataGrid::of(search_orders).column(Order::code).filter().column(Order::total).sum(),
    )
}

No HTML, CSS or JavaScript to write. The grid pages, filters and sums on the server, and the same action answers the API and AI agents.

In the box

DataGrid Virtual scrolling over a million rows, typed filters, totals, grouping, inline editing, CSV and Excel export.
Components Layouts, date and time pickers, controls, lines editor, tree, charts, command palette, menus, docking panels.
Security Permissions, guards, interceptors, field audit, read audit, masking, rate limits, IP rules.
Data PostgreSQL and Oracle 11g or later, keyset cursors, transactions, streams, migrations.
Sign in OIDC, JWT, MFA, passkeys and an optional accounts system with ready screens.
Runs light A 43 MB container that idles at about 3 MiB of memory, or one desktop exe.