Language

Honest AI Lang is the text format consumed by Renderer. A program declares a component tree, values, and actions. You can write it by hand to test an integration, then give the same component signatures to a model.

Every rendered program needs a root statement. Use Stack for the application library:

root = Stack([heading, details])
heading = InlineHeader("Example checklist", "Before departure")
details = Card([TextContent("Confirm the booking and pack your charger.")])

The renderer follows root to its children. heading and details are references to other statements; they can be declared after the statement that uses them. Inline children such as TextContent(...) are allowed too.

In a chat reply, a program is a fenced honestai block inside Markdown. It uses the chat library, whose Stack root takes only children:

root = Stack([Callout("info", "Example checklist", "Confirm the booking and pack your charger.")])

See the chat reply format.

Send the matching spec.json or chat-spec.json to your prompt generator. The root is part of that contract, not a layout preference inferred by the renderer.

Component calls use positional arguments. Read each signature from left to right. TextContent(text: string, size?: …) takes text first and size second:

root = Stack([TextContent("Example note", "small")])

An argument marked ? is optional. To supply a later argument while omitting an earlier optional one, put null in the earlier position. Do not replace positional arguments with a props object unless that argument's schema is itself an object.

Strings use double quotes. Arrays use square brackets. Objects use braces. Use \n inside a string for a line break, such as a Markdown paragraph break. Names and enum values must match the library exactly; TextContent and textcontent are different names.

The component library reference lists registered names. Its generated JSON spec contains the complete signatures, allowed child types, and argument schemas.

A $ prefix marks a value that can change after the program renders. An ordinary statement names a component, a query, or an expression. For example:

$quantity = 1
root = Stack([TextContent("Example count: " + $quantity), button])
button = Button("Add one", @Set($quantity, $quantity + 1), "primary")

@Set changes the reactive value. The text reads that value, so it updates without a model request. @Reset($quantity) restores the declared default. See State and forms for a live example and snapshot restoration.

Only props marked as reactive bindings accept a variable for two-way field updates. The library signature identifies those slots; passing a variable to an arbitrary prop does not turn that prop into an editable field.

Action steps have an @ prefix. A button can accept one step or an Action([...]) sequence.

@ToAssistant and @OpenUrl reach your onAction callback. The chat shell provides default handling for those two actions; a standalone renderer needs your handler. @Run uses the registered tool implementation. See Data and tools before adding reads or writes.

The renderer accepts a bare program or a stored message built by the local protocol helpers:

import { buildMessage } from "@/lib/generative"

export const stored = buildMessage({
  content: 'root = Stack([TextContent("Saved example")])',
})

The stored-message protocol uses ]]>honestai: markers around its sections. Use buildMessage and parseMessage rather than constructing marker boundaries by hand. A code fence uses the language name honestai; a bare program is enough for a direct Renderer response.

An unknown component usually means a name is wrong or the selected library does not register it. Missing arguments mean the call does not follow its signature. Unresolved references mean a named statement was used but never defined. An incomplete statement may be normal while tokens are still arriving; it is an error to resolve after generation ends.

Try these failures in the diagnostics demo. If the source parses but a data request fails, retry the request through the query runtime rather than rewriting the program.