unstructured
91 versions indexed · latest 0.27.37 on Nov 6, 2026
202656
fix(lang): keep the languages a caller passes on short ASCII text · fix(pdf): extract vector-heavy PDFs with the fast strategy instead of returning nothing
chore: start 0.27.36-dev after 0.27.35 release · fix(xml): keep text that runs through child elements · fix(pptx): filter off-slide shapes by how much of the box is visible
chore: replace per-PR changelog edits with generated release notes · fix(utils): compare bounding boxes for elements with no page number · fix(chunking): measure combine thresholds in tokens when max_tokens is set
perf: Free decode buffers sooner when restoring original elements · perf: Free image buffers sooner during figure extraction · fix: honor encoding= in partition_json and partition_ndjson
fix(chunking): don't repeat a very large split element in every chunk · fix(deps): bump certifi to 2026.10.19
fix: label numbered headings without making them list items, and skip section-break paragraphs · fix(partition): read spooled uploads in place instead of copying them
fix(email): keep non-ASCII sender and recipient names readable
fix(xlsx): read .xls files whose Workbook stream ends in padding
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
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(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
This release rolls up all changes since 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.
fix(html): extract definition lists instead of discarding them
fix(chunking): retain table rowspans across cell splits
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 DOCX text_as_html duplicating merged-cell text instead of colspan/rowspan
chore(deps): bump claude-code-action to v1 · Added Sign up link for Transform and removed pricing · fix: stop treating every hyphen as a bullet delimiter in text partitioning
fix: support core metadata 2.5 publishing
test: stop GLOBAL_WORKING_DIR tests from disturbing other xdist workers · chore: restore telemetry by default · fix(security): bound quadratic array-stream decoding in is_pdf_too_complex (SEC-146)
Terminology update: Platform -> Pipelines in README · perf(html): use indexed parent lookup · feat: add lazy chunking entry points
Partition arbitrary JSON and NDJSON: partition_json() and partition_ndjson() now handle any valid JSON/NDJSON payload, not just serialized Unstructured output.
fix: sanitize v2 HTML output to prevent stored XSS (GHSA-v5mq-3xhg-98m9)
feat: derive category_depth from heading level in the v2 (ontology) HTML parser · fix(ci): stabilize partition benchmark with a rolling-median baseline
feat: extract filled AcroForm field text in PDF partitioning
fix: stop decimating embedded text on dense PDF pages · fix: keep extracted text aligned with rotated PDF page images in hi_res · feat: add enrichment origins metadata field
fix(hi_res): recover text inside PDF figure overlays
fix: rename isolate_tables chunking option to isolate_table
feat: add option for table chunking
fix: handle text too long for spacy issue
fix: chunking dropping table content
fix: ndjson file type detection
Reject oversized PDF renders before bitmap allocation · feat(cli): add unstructured doctor diagnostics command · feat: Track the table extraction method
fix: first table chunk preserve col/row span
Replace PyPI opencv wheels with ffmpeg-free builds in Docker image: After uv sync, the Dockerfile now substitutes all PyPI opencv-python variants with a source-built opencv-contrib-python-headless wheel compiled with…
feat: add option to skip table chunking
Fix fixtures update CI to regenerate markdown · fix(deps): upgrade vulnerable transitive dependencies [security][bot] · feat: add GHA workflow to build opencv wheels without ffmpeg
OpenCV Python contrib-headless wheels built from source with WITH_FFMPEG=OFF.
fix(chunking): preserve semantic headers in carried table chunks · feat: add page number support to v1 html partition
Formula markdown export (element_to_md / elements_to_md): New keyword-only formula_markdown_style ('auto', 'display_math', 'plain'; default 'auto'). In 'auto', display math ($$ ...
mem: exclude unused spaCy pipeline components to reduce model memory · fix: pdfminer drops extractable text
fix(chunking): preserve nested table structure in reconstruction · Replace lazyproperty with functools.cached_property · mem: reduce PaddleOCR rec_batch_num from 6 to 1 · fix: isolate Table elements in pre-chunks
fix(deps): Update security updates [SECURITY][bot] · fix: Self-contained script for version extraction in release CI
feat: add create_file_from_elements() to re-create document files from elements · Bump dependencies · fix: avoid O(N²) re-scanning in patch_current_chars_with_render_mode · add check if libmagic fails
feat: custom fallback for language detection · Add Github action for time regressions · fix: relax lower bound for pdfminer.six
Replace NLTK with spaCy to remediate CVE-2025-14009: NLTK's downloader uses zipfile.extractall() without path validation, enabling RCE via malicious packages (CVSS 10.0, no patch available).
fix: set max decompressed size for elements JSON · fix: update depdencies
Automate pypi publishing · fix: remove duplicate characters caused by fake bold rendering in PDFs · Improve fast partition cold start · fix: gracefully handle invalide html string during chunking
Add Python 3.11 and 3.13 support · Use bigger runner to publish images
feat: add group_elements_by_parent_id utility function · Preserve newlines in Table and TableChunk elements during PDF partitioning · Fix: make pdf image dpi consistent · chore: bump dependencies for 0.18.34
feat: put pdfium call behind a threadlock
Feat: patch pdfminer and use rendermode to detect invisible text · fix: add EN DASH to UNICODE_BULLETS for clean_bullets · fix: fix version number · enhancement: render pdfs with pdfium
Optimize clean_extra_whitespace_with_index_run (codeflash) · Optimize recursive_xy_cut_swapped (codeflash) · Optimize DocxPartitioner.parse_category_depth_by_style_name (codeflash)
Comment no-ops in zoom_image (codeflash) · Fix an issue where elements with partially filled extracted text are marked as extracted
Pin deltalake<1.3.0 to fix ARM64 Docker builds (1.3.0 missing Linux ARM64 wheels)
202533
Optimize OCRAgentTesseract.extract_word_from_hocr (codeflash)
fix(deps): Bump fonttools to address cve
Update save_elements unit test to check crop box padding behavior
Improve the VoyageAI integration · Add voyage-context-3 support · Flag extracted elements as such in the metadata for downstream use
Prevent path traversal in email MSG attachment filenames Fixed a security vulnerability (GHSA-gm8q-m8mv-jj5m) where malicious attachment filenames containing path traversal sequences could write files outside the…
Setup Codeflash Github Actions to optimize all future code · fix: update deps to resolve cve · ⚡️ Speed up function group_broken_paragraphs by 30% · ⚡️ Speed up method ElementHtml.get_children_html by 234%
Speed up function sentence_count by 59% (codeflash) · Speed up function check_for_nltk_package by 111% (codeflash) · Speed up function under_non_alpha_ratio by 76% (codeflash)
Parse a wider variety of date formats in email headers The partition_email function is now more robust to non-standard date formats, including ISO-8601 dates with 'Z' suffixes.
Prevent large file content in encoding exceptions Replace UnicodeDecodeError with UnprocessableEntityError in encoding detection to avoid storing entire file content in exception objects, which can cause issues in…
add '|' as a delimiter in csv files · feat: map tags by type + add coverage · chore: switch to charset normalizer · bump version and release
Add OCR_AGENT_CACHE_SIZE environment variable Added configurable cache size for OCR agents to control memory usage.
Convert elements to markdown for output Added function to convert elements to markdown format for easy viewing.
Add language detection for PDFs Add document and element level language detection to PDFs.
Improved epub partition errors EPUB partition will now produce new type of error on unprocessable files.
fix(partition, csv): increase csv field limit
chore: bump pillow to address a CVE
fix [NEX-49] : Fix TypeError for empty HTML content · fix: add try/except wrap over row.cells to failproof tc grid_offset · fix: xml processing not escaped · fix: update md to reads umlauts on non-utf-8 files
Add DocumentData element type This is helpful in scenarios where there is large data that does not make sense to represent across each element in the document.
Add image_url of images in html partitioner tags with non-data content include a new image_url metadata field with the content of the src attribute.
feat: include images when partitioning html · fix: pass extract image args to all partitioners · feat: allow passing down of ocr agent and table agent · Feat/remove reference of PageLayout.elements
Fixes filetype detection for jsons passed as byte streams - Now it prioritizes magic mimetype prediction over file extension when detecting filetypes
Support dynamic partitioner file type registration. Use create_file_type to create new file type that can be handled
Fixes detect_filetype when SpooledTemporaryFile is passed. Previously some random name would get assigned to the file and the function raised error.
Fix open CVES in and bump dependencies
Use password to load PDF with all modes
Fix a security issue where rst and org files could read files in the local filesystem. Certain filetypes could 'include' or 'import' local files into their content, allowing partitioning of arbitrary files from the…
Fix a bug where table extraction is skipped when it shouldn't. Pages with just one table as its content or starts with a table misses table extraction. The routing logic is now fixed.
Refactoring the VoyageAI integration to use voyageai package directly, allowing extra features.
Vectorize layout (inferred, extracted, and OCR) data structure Using np.ndarray to store a group of layout elements or text regions instead of using a list of objects.
Update unstructured-inference to 0.8.6 in requirements which removed layoutparser dependency libs · Update pdfminer-six to 20240706
Fix an issue with multiple values for infer_table_structure when paritioning email with image attachements the kwarg calls into partition to partition the image already contains infer_table_structure.
Add character-level filtering for tesseract output. It is controllable via TESSERACT_CHARACTER_CONFIDENCE_THRESHOLD environment variable.
Prepare auto-partitioning for pluggable partitioners. Move toward a uniform partitioner call signature so a custom or override partitioner can be registered without code changes. · Add NDJSON file type support.
20242
Enhance quote standardization tests with additional Unicode scenarios
Fix original file doctype detection from cct converted file paths for metrics calculation.