ax

Call Signature

function ax<T, ThoughtKey>(signature: T, options?: Readonly<AxAIServiceOptions & object & object>): AxGen<ParseSignature<T>["inputs"], ParseSignature<T>["outputs"] & string extends ThoughtKey ? object : { [P in string]?: string }>;

Defined in: https://github.com/ax-llm/ax/blob/05ff5bd88d050f7ba85a3fcc6eb0ed2975ad7d51/src/ax/dsp/template.ts#L110

Creates a type-safe AI generator from a signature string or AxSignature object.

This is the primary way to define AI-powered functions in Ax. The signature string declares input and output fields with their types, which are then used to generate prompts and parse responses.

Signature String Format:

"inputField1: type, inputField2: type -> outputField1: type, outputField2: type"

Supported Field Types:

string - Text content (default if no type specified)
number - Numeric values (integers or floats)
boolean - True/false values
json - Arbitrary JSON objects
date - Date in YYYY-MM-DD format
datetime - ISO 8601 datetime
code - Code blocks (preserves formatting)
image - Image input (for multimodal models)
audio - Audio input
class - Classification with predefined options: class(option1, option2, option3)

Type Modifiers:

[] suffix - Array of values: tags: string[]
? suffix - Optional field: context?: string
! prefix - Internal field (hidden from output): !reasoning: string

Field Descriptions: Add descriptions after the type using a string literal:

"question: string 'The user question' -> answer: string 'A helpful response'"

Type Parameters

Type Parameter	Default type
`T` extends `string`	-
`ThoughtKey` extends `string`	`"thought"`

Parameters

Parameter	Type	Description
`signature`	`T`	Either a signature string or a pre-built AxSignature object
`options?`	`Readonly`<`AxAIServiceOptions` & `object` & `object`>	Optional configuration for the generator

Returns

AxGen<ParseSignature<T>["inputs"], ParseSignature<T>["outputs"] & string extends ThoughtKey ? object : { [P in string]?: string }>

An AxGen instance that can be executed with .forward(ai, inputs)

Examples

const qa = ax('question: string -> answer: string');
const result = await qa.forward(ai, { question: 'What is TypeScript?' });
console.log(result.answer);

const classifier = ax('text: string -> sentiment: class(positive, negative, neutral)');
const result = await classifier.forward(ai, { text: 'I love this!' });
console.log(result.sentiment); // 'positive'

const extractor = ax(`
  document: string ->
  summary: string,
  keywords: string[],
  wordCount: number
`);
const result = await extractor.forward(ai, { document: longText });

const solver = ax('problem: string -> solution: string', {
  thoughtFieldName: 'reasoning'
});
// Enable thinking in forward options to get step-by-step reasoning

const agent = ax('query: string -> response: string');
const result = await agent.forward(ai, { query: 'What is 25 * 4?' }, {
  functions: [{
    name: 'calculate',
    description: 'Perform math calculations',
    parameters: { type: 'object', properties: { expression: { type: 'string' } } },
    func: ({ expression }) => eval(expression)
  }]
});