Changelog

Set advancedOptions.htmlSourceAnchors to true in your parse config to anchor each block parsed from an HTML file to the exact byte range it came from. Anchored blocks carry a source object (type, byteOffset, byteLength) in place of polygon/boundingBox, and the run reports the encoding those offsets are measured against in output.metadata.sourceEncoding. Add advancedOptions.returnSourceNodes for word-level anchors under output.source.nodes, the text-format counterpart to returnOcr.words.

Available on API version 2026-02-09, in Extend Studio, and with any parse engine and version. HTML files must be UTF-8 or Windows-1252 encoded; otherwise the run fails with the new UNSUPPORTED_FILE_ENCODING failure reason.

SDK note: because anchored blocks omit polygon and boundingBox, both fields are now optional in the SDK block types starting with extend-ai 2.5.0 (TypeScript), extend-ai 1.21.0 (Python), extend-java-sdk 1.22.0, and extend-go-sdk 0.9.0. Upgrading Java or TypeScript may require wrapping existing geometry access in an Optional or null check; Python and Go are unaffected at runtime. Responses for non-HTML parses, and for HTML parses without this option, are unchanged.

See HTML source anchors and the configuration reference.