Quick Answer

Extraction APIs let you generate structured outputs using schemas and replay-aware plans so AI workflows stay stable over time.

Extraction

extract<T>(options: ExtractOptions): Promise<T>

Extract structured data from the page.

schema describes the output shape, not just selector bindings. You can use semantic field hints such as "string" and arrays of objects for AI extraction, or explicit bindings such as { element: 3 }, { element: 5, attribute: "href" }, { selector: ".price" }, and { source: "current_url" } for deterministic extraction. Use prompt to describe relationship/fallback rules.

Relative URL-like attributes extracted from accessible iframe documents are resolved against the iframe document base URL. Main-frame extraction keeps the existing raw relative attribute behavior.

const images = await opensteer.extract({
    description: 'article images with captions and credits',
    prompt:
        'For each image, return the image URL, alt text, caption, and credit. Prefer caption and credit from the same figure. If missing, look at sibling text, then parent/container text, then nearby alt/data-* attributes.',
    schema: {
        images: [
            {
                imageUrl: 'string',
                alt: 'string',
                caption: 'string',
                credit: 'string',
            },
        ],
    },
})

const data = await opensteer.extract({
    description: 'product-info',
    schema: {
        title: { element: 3 },
        price: { element: 5 },
        url: { source: 'current_url' },
    },
})

For explicit array bindings, include multiple representative items so Opensteer can infer the repeating pattern. For semantic extraction, a single representative object shape is enough.

extractFromPlan<T>(options: ExtractFromPlanOptions): Promise<ExtractionRunResult<T>>

Extract data from a pre-built extraction plan with explicit field mappings and element paths.

Extract Options

interface ExtractOptions extends BaseActionOptions {
    schema?: ExtractSchema
    prompt?: string
    snapshot?: SnapshotOptions
}

interface ExtractFromPlanOptions {
    description?: string
    schema: ExtractSchema
    plan: ExtractionPlan
}

interface ExtractSchema {
    [key: string]: ExtractSchemaValue
}

type ExtractSchemaValue =
    | ExtractSchemaField
    | string
    | number
    | boolean
    | null
    | ExtractSchema
    | ExtractSchema[]

interface ExtractSchemaField {
    element?: number
    selector?: string
    attribute?: string
    source?: 'current_url'
}

Covered Features

• sdk:extract
• sdk:extractFromPlan

Related Docs

• OpenSteer Actions API Reference
• Agent + CUA
• Errors and Diagnostics
• OpenSteer SDK Lifecycle Reference