unstructured 0.27.10
- fix(html): extract definition lists instead of discarding them
Also shipped Sep 27, 2026Unstructured in September 2026 →
Ember-1 from Fireworks now available on AI Gateway
Ember-1 from Fireworks is now available on AI Gateway . Ember-1 is a research preview reasoning model built on Kimi K3 for coding and agentic workflows. Fireworks reports approximately 40% fewer generated tokens than Kimi K3 at comparable quality across its evaluations.
On-demand rule execution and Conversation Metadata now generally available with Twilio Conversations
Today, we are launching two new capabilities with Twilio Conversations – On-demand rule execution for Conversation Intelligence and Conversation metadata for Conversation Orchestrator
Usage tracking adds session metadata and token details (Public Preview)
The system.ai_gateway.usage system table now includes coding-agent session metadata and invocation source. New token-detail fields capture tokens written to the 5-minute and 1-hour prompt caches, along with file and web search counts. See Track model usage .
feldera 0.357.0
- [adapters] Take over a checkpoint bucket at startup with take_bucket_ownership [SQL] Do not merge two GC operators if their control inputs are not 'equivalent' [SQL] Fix implementation of WATERLINE for LEFT JOIN [SQL] Fix incorrect WATERLINE computation for ASOF JOIN key (currently unreachable) [SQL] Fix HOP/TUMBLE implementation to handle dates before...
Workflows, Workers - Call Workflows declared in `exports` through `ctx.exports`
A Worker can now call the Workflows it declares in the exports field of its Wrangler configuration through ctx.exports . You no longer need a workflows binding to call a Workflow from the Worker that defines it.
More from unstructuredEvery unstructured release →
unstructured 0.27.9
- fix(chunking): retain table rowspans across cell splits
unstructured 0.27.8
- fix: partition multi-section DOCX files in linear time - fix: ignore processing instructions in HTML partitioning - ci: remove release version alert - fix: tolerate non-UTF-8 soffice output in doc/ppt conversion - fix: recognize single-block list items as ListItem - perf(nlp): run the spaCy pipeline once per distinct text - fix(html): preserve table...
unstructured 0.27.16
This release rolls up all changes since 0.27.10; versions 0.27.11–0.27.15 were not published to PyPI individually. Fixes Reject images whose frames decode to too many pixels instead of exhausting memory.
unstructured 0.27.22
This release rolls up all changes since 0.27.16. Fixes - Faster chunk serialization with stable source-element IDs. Avoid copying orig_elements twice during serialization. IDs minted during serialization now remain stable across calls, including inside orig_elements; explicit and hash-derived IDs are unchanged.
unstructured 0.27.25
- fix(pptx): no page break before the first slide when starting_page_number > 1 - fix(chunking): reject a negative overlap instead of corrupting text - feat(image): partition WEBP images like PNG and JPEG
unstructured 0.27.6
- Fix DOCX text_as_html duplicating merged-cell text instead of colspan/rowspan
unstructured 0.27.27
- fix(json): preserve rehydrated element identity - fix: keep overlap category names from element 100 on - fix: normalize None text to empty string in Text elements - fix: stop silently returning mojibake from encoding detection fix: partition_email raises KeyError on a multipart/ attachment - fix: preserve MathML content in HTML partitioning
unstructured 0.27.28
- feat: opt-in langdetect-rs backend for partition language detection - fix(html): keep text inside unrecognized elements - fix(html): keep table captions in Table text and text_as_html - perf(docx): resolve each paragraph style once per document (5.3 s -> 1.0 s outside spaCy) - fix(xlsx): collapse empty-row runs instead of rejecting far-away cells -...