v2.128.0 v2.129.0
tanchongmin/strictjsonSep 16, 2026 → Sep 18, 20262 days between releases
What's New
Adds ambient S3 credentials, chart extraction refactor, and improves Asciidoc/HTML/CSV parsing.
New Features
- Refactor chart extraction with generic engine options
- Allow ambient credentials for S3
- Support colon, dash, bracket and parenthesized numbering markers
- Decode text documents that are not UTF-8
- Propagate token usage from VLM engine output
Improvements
- Normalize CRLF and lone CR in message bodies
- More improvements for Chandra OCR parsing
- Default a zero colspan or rowspan to 1 in HTML tables
Bug Fixes
- Recognise a rowspan-only cell specifier in Asciidoc
- Wrap strict quote errors as DocumentLoadError in CSV
- More improvements for Chandra OCR parsing
- Default a zero colspan or rowspan to 1 in HTML tables
Upgrade Notes
Chart extraction now uses generic engine options; verify your configuration if you rely on specific extraction parameters. CSV parsing now wraps strict quote errors in DocumentLoadError for better error handling. Ambient S3 credentials are now supported automatically.
The Two Releases
Sep 16, 2026· 1 month ago
Adds minimal AFP document support and NVIDIA Nemotron Parse v2 VLM. Refactors the PDF backend by removing pypdfium.
Full release notesSep 18, 2026· 1 month ago
Refactors chart extraction with generic engine options and adds ambient S3 credentials support. Improves parsing for Asciidoc, HTML, CSV, and email formats.
Full release notes