Process Document with Claude OCR (via ellmer) — claude_ocr_process

Processes a local PDF or image with Claude's document understanding using ellmer for structured output, returning Tensorlake-compatible page data. The output schema is enforced with an ellmer type specification, so the model reliably returns pages, tables, and other elements.

Usage

claude_ocr_process_file(
  file_path,
  api_key = Sys.getenv("ANTHROPIC_API_KEY"),
  model = "claude-opus-4-6",
  max_tokens = 16000,
  extraction_prompt = NULL,
  pages = NULL
)

Arguments

file_path: Character string. Path to a local PDF, PNG, JPEG, GIF, or WEBP file.
api_key: Character string. The Anthropic API key. Default retrieves from environment variable "ANTHROPIC_API_KEY".
model: Character string. The Claude model to use. Default is "claude-opus-4-6".
max_tokens: Integer. Maximum tokens in the response. Default is 16000.
extraction_prompt: Character string. Custom task prompt. If NULL, uses a default prompt (the output structure itself is enforced by the type schema).
pages: Integer vector. Accepted for interface parity; Claude reads the whole document, so page selection is applied later in claude_extract_pages. Default NULL.

Value

A list with structured_output$pages: a list with one entry per page (page_number, page_header, section_header, text, tables, other).

Author

Nathan C. Layman