AlterLab icon

AlterLab

Scrape any website with anti-bot bypass, JS rendering, structured extraction, OCR, and more

Actions2

Extract

AI-generated

Summary

Extract structured or transformed data from supplied HTML, Markdown, or text without fetching a URL. Configure output formats (which now include raw unmodified body), an optional JSON schema, profile or template (mutually exclusive), prompt, model, and provider. Provide source URL context and evidence options, and control cache behavior with new force refresh support. The node posts each input item to AlterLab's extract API, supporting asynchronous submission and returning job data when requested.

Inputs

  • Content (required) — Raw HTML, Markdown, or text to process; the UI documents a 500,000-character maximum.
  • Content Type — Identifies the supplied content as HTML (default), Markdown, or text.
  • Extract Options → Formats — Choose output formats including Content (AI/KB), HTML, JSON (default), JSON V2 section tree, Markdown, chunked RAG, Raw (unmodified body), or text. The API response's selected formats are flattened into output fields.
  • Extract Options → Extraction Schema (JSON) — JSON schema for structured extraction; invalid JSON causes an item error. Mutually exclusive with auto extraction and templates.
  • Extract Options → Extraction Profile — Optional predefined schema template: None, Auto, Article, Directory Listing, Ecommerce Homepage, Event, FAQ, Job Posting, Product, or Recipe. Mutually exclusive with Extraction Template except Auto.
  • Extract Options → Extraction Template — Optional shorthand alias for Extraction Profile (None, Auto, Article, Directory Listing, Ecommerce Homepage, Event, FAQ, Job Posting, Product, Recipe). Mutually exclusive with a non-Auto Extraction Profile.
  • Extract Options → Extraction Prompt — Natural-language instructions for the LLM.
  • Extract Options → Extraction Model — Optional provider-specific model override for this request.
  • Extract Options → Extraction Provider — Choose OpenAI, Anthropic, OpenRouter, or Groq; blank selects the most recently used registered BYOK key.
  • Extract Options → Source URL — Optional original URL supplied for context only; it is not fetched.
  • Extract Options → Evidence — Enable provenance or evidence for extracted fields.
  • Extract Options → Cache Control — Auto (default) returns a cached extraction when available, otherwise calls the LLM and caches the result; Skip always calls the LLM and stores the result; Only returns a cached result or an HTTP 404.
  • Extract Options → Cache TTL (Seconds) — Cache duration from 1 to 86,400 seconds; default is 3,600 seconds.
  • Extract Options → Force Refresh — Whether to bypass the cache and fetch fresh content even when Cache is enabled.

Output shape

One item per input item, each containing an extraction result object.

The returned item for each input contains extraction metadata including extractId, creditsUsed, modelUsed, extractionMethod, extractionProfile, contentSizeChars, cacheHit, and extractionMetadata when present. The requested format outputs are provided in flattened fields named text, markdown, html, json, jsonV2, rag, content, and raw (unmodified response body) when returned. Advanced outputs include screenshotUrl, pdfUrl, ocrResults, rawHtml, and detailed billing breakdown with BYOP/BYOS and surcharge fields. The node supports asynchronous extract submission with job ID and poll URL outputs. API errors, including invalid schema JSON and cache-only misses, fail the item unless Continue On Fail is enabled.

Links

Discussion