v2026.03.18

What's Changed

• Pulled in pyzotero typing fix by @jamesbraza in #1319
• Loosen typing of actions to Message by @sidnarayanan in #1318

Full Changelog: v2026.03.12...v2026.03.18

v2026.03.12

What's Changed

• chore(deps): lock file maintenance by @renovate[bot] in #1315
• chore(deps): update actions/download-artifact action to v8 by @renovate[bot] in #1314
• Pulling in docling-core, docling, pymupdf fixes by @jamesbraza in #1313
• Add "et al." to invalid citation examples in prompt by @jamesbraza in #1316

Full Changelog: v2026.03.03...v2026.03.12

v2026.03.03

What's Changed

• Upstreamed npm deps caching by @jamesbraza in #1308
• Fixing JSON schema export of Settings by @jamesbraza in #1309
• Fixing flaky PaSa figure 1 read assertions by @jamesbraza in #1312
• Fixed CMYK images crashing PNG encoding in PyPDF reader by @jamesbraza in #1311

Full Changelog: v2026.02.27...v2026.03.03

v2026.02.27

What's Changed

• Downpinning docling-parse, PyMuPDF, pyzotero, litellm bugs by @jamesbraza in #1295
• chore(deps): lock file maintenance by @renovate[bot] in #1292
• Respecting custom index settings name by @jamesbraza in #1296
• Adding prompt caching by @jamesbraza in #1293
• Fixing Semantic Scholar crash on max over empty list by @jamesbraza in #1297
• Fixed PyPDF reader's failure to handle non-PNG data without pdfplumber by @jamesbraza in #1298
• Handling LDP Agent incorrectly giving Message by @jamesbraza in #1299
• Caching npm deps in lint CI by @jamesbraza in #1304
• chore(deps): lock file maintenance by @renovate[bot] in #1303
• Re-loosened test_timeout_resilience after LiteLLM fix by @jamesbraza in #1305
• Disabling cache reads assertion for Google Gemini by @jamesbraza in #1306
• Multiprocessing support + optimization for nemotron-parse reader by @jamesbraza in #1300
• Fix acronym-led citation docname ingest failures by @AmT42 in #1302
• Supporting signed GCS links in ParsedMedia by @jamesbraza in #1307

New Contributors

• @AmT42 made their first contribution in #1302

Full Changelog: v2026.02.16...v2026.02.27

v2026.02.16

What's Changed

• Added del to save memory during reading by @jamesbraza in #1265
• More nemotron-parse links in README.md by @jamesbraza in #1267
• chore(deps): lock file maintenance by @renovate[bot] in #1268
• Validating ParsedMedia.data is not empty by @jamesbraza in #1269
• Retrying nemotron-parse API calls receiving 408 timeouts by @jamesbraza in #1276
• Fixing 60-sec wait for retrying nemotron-parse API calls' 408 timeouts by @jamesbraza in #1277
• Updating journal_quality.csv from script by @jamesbraza in #1278
• Logging tool name in NemotronBBoxError/NemotronLengthError by @jamesbraza in #1279
• Non-destructive retrying on nemotron-parse API by @jamesbraza in #1281
• Per-page failover parser for nemotron-parse by @jamesbraza in #1280
• Fixing newly-added journal quality 4 causing KeyError by @jamesbraza in #1282
• Pulled in UV_VENV_CLEAR for uv==0.10.0 break by @jamesbraza in #1285
• Fixing test_parse_office_doc by modernizing Gemini model by @jamesbraza in #1286
• Multiprocessing support for PyMuPDF full_page mode by @jamesbraza in #1284
• Fixing failing test_equations[docling] by caching Docling models before pytest by @jamesbraza in #1287
• Fixing dockey/doc_id mismatch when no metadata is found by @jamesbraza in #1288
• chore(deps): update suzuki-shunsuke/github-action-renovate-config-validator action to v2 by @renovate[bot] in #1291
• chore(deps): update pre-commit hook psf/black-pre-commit-mirror to v26 by @renovate[bot] in #1290

Full Changelog: v2026.01.05...v2026.02.16

v2026.01.05

What's Changed

• Fixing crash from bad PDF read with PyMuPDF by @jamesbraza in #1259
• Fixing crash from bad PDF read with PyPDF by @jamesbraza in #1260
• chore(deps): lock file maintenance by @renovate[bot] in #1263
• Not having ID break index cache key across Python invocations by @jamesbraza in #1262
• Refreshing test_equations[docling] VCR cassette by @jamesbraza in #1264

Full Changelog: v2025.12.23...v2026.01.05

v2025.12.23

What's Changed

• Unsilenced flaky test_duplicate_media_context_creation by @jamesbraza in #1252
• Modernizing bundled configs by @jamesbraza in #1251
• Moving nemotron-parse to failover to reinvent markdown_bbox on length error by @jamesbraza in #1256
• Refreshing settings table and tutorial by @jamesbraza in #1255

Full Changelog: v2025.12.19...v2025.12.23

v2025.12.19

What's Changed

• Disabling parallel CI jobs' fast fail by @jamesbraza in #1248
• Adding some pytest.mark.flaky by @jamesbraza in #1249
• Added IoU-based merging to nemtron-parse by @jamesbraza in #1246
• Fixed bug in PyPDF reader where one can't avoid pdfplumber by @jamesbraza in #1247

Full Changelog: v2025.12.17...v2025.12.19

v2025.12.17

Summary

The last four months since version 5.29.1 have seen many changes:

• New modalities: tables, figures, non-English languages, math equations
• More and better readers
  • Two new model-based PDF readers: Docling and Nvidia nemotron-parse
  • All PDF readers now can parse images and tables, report page numbers, support DPI
  • A reader for Microsoft Office data types
• Multimodal contextual summarization
  • Media objects are also passed to the summary_llm during creation
  • Media objects' embedding space is enhanced using an enrichment_llm prompt
• Simpler and performant HTTP stack
  • Consolidation from aiohttp and httpx to just httpx
  • Integration with httpx-aiohttp for performance
• Context relevance is simplified and some assumptions were removed
• Many minor features such as retrying Context creation upon invalid JSON, compatibility with fall 2025's frontier LLMs, and improved prompt templates
• Multiple fixes in metadata processing via Semantic Scholar and OpenAlex, and metadata processing (e.g. incorrectly inferring identical document IDs for main text and SI)
• Completed the deprecations accrued over the past year

What's Changed

• Fixing gen_answer failover leaving raw_answer blank by @jamesbraza in #1077
• Image reader and image support in gather_evidence by @jamesbraza in #1046
• Multimodal PDF support by @jamesbraza in #1047
• Documenting DocMetadataTask/MetadataProvider by @jamesbraza in #1050
• chore(deps): lock file maintenance by @renovate[bot] in #1075
• Fixed LDP erroneous v33 tag by @jamesbraza in #1076
• Consolidating on httpx by @jamesbraza in #1062
• S2 metadata retrieval: fallback to exact title match if authors do not match by @sidnarayanan in #1078
• chore(deps): update pre-commit hook pre-commit/pre-commit-hooks to v6 by @renovate[bot] in #1080
• chore(deps): update actions/setup-python action to v6 by @renovate[bot] in #1081
• Retrying Context creation once by @jamesbraza in #1083
• Expanding llm_parse_json and removing extract_score assumptions by @jamesbraza in #1082
• Fixing flaky test_title_search via client ordering by @jamesbraza in #1084
• Newer tooling and deps by @jamesbraza in #1085
• Fixed type hint/docstring on DocMetadataClient.metadata_clients by @jamesbraza in #1086
• chore(deps): lock file maintenance by @renovate[bot] in #1087
• Filtering out null byte from table ParsedMedia.text by @jamesbraza in #1088
• Context id updates by @mskarlin in #1089
• Removed other from doc details for user by @whitead in #1092
• Removing inplace modification by @whitead in #1093
• Enforcing DocDetails.tzinfo to be UTC by @jamesbraza in #1094
• Migrating httpx.AsyncClient to httpx_aiohttp.HttpxAiohttpClient by @jamesbraza in #1099
• Adopting prek over pre-commit, setup-uv's python-version by @jamesbraza in #1098
• Gather evidence fails gracefully on timeout by @sidnarayanan in #1103
• Upstreaming prompt change by @whitead in #1100
• gather_evidence tool response filtering irrelevant Context, with updated threshold of 0 by @jamesbraza in #1106
• Fixing used_images usage in summary_json_system_prompt by @jamesbraza in #1109
• Removed actions/setup-python from pre-commit by @jamesbraza in #1102
• Newer pre-commit deps, to test pre-commit CI works by @jamesbraza in #1110
• Dropping irrelevant contexts by @jamesbraza in #1113
• Catch safety refusals and other BadRequestErrors in gather evidence by @sidnarayanan in #1114
• Exposing DOC_DETAILS_OTHERS_TO_KEEP for subclassers by @jamesbraza in #1115
• Re-adding refurb by @jamesbraza in #1117
• Adding missing py.typed to reader packages by @jamesbraza in #1118
• Fixing uv cache not existing by @jamesbraza in #1120
• chore(deps): lock file maintenance by @renovate[bot] in #1119
• Fixing KeyError crash when PDF reader misses a page by @jamesbraza in #1122
• Fixing Settings.get_index_name being the same for different PDF parsers by @jamesbraza in #1125
• Ensuring 404 PDF is not parsed into texts by @jamesbraza in #1126
• Updating LLM descriptions in Settings by @jamesbraza in #1130
• Cleaning up tiktoken usage in readers by @jamesbraza in #1131
• Pulling in BAIPP's fix that ignores UV_PYTHON by @jamesbraza in #1133
• chore(deps): update astral-sh/setup-uv action to v7 by @renovate[bot] in #1134
• chore(deps): lock file maintenance by @renovate[bot] in #1135
• Refactored ParsedMetadata/ChunkMetadata to reflect all options by @jamesbraza in #1132
• Pulling in LMI's image utils by @jamesbraza in #1129
• Created docling PDF reader by @jamesbraza in #1121
• chore(deps): update astral-sh/setup-uv action to v7 by @renovate[bot] in #1136
• Pinning fhlmi min version by @jamesbraza in #1137
• Ensuring citation peeks aren't multimodal by @jamesbraza in #1138
• Not inferring title/DOI every docs_fixture construction by @jamesbraza in #1139
• Making search tools concurrent by @sidnarayanan in #1141
• Deprecating Docs.delete's name arg by @jamesbraza in #1142
• Fixing flaky Docling tests due to filesystem race conditions by @jamesbraza in #1144
• Removed unnecessary fhlmi pinning by @jamesbraza in #1146
• Expanding temperature autoset to 1 for GPT-5 by @jamesbraza in #1145
• chore(deps): lock file maintenance by @renovate[bot] in #1148
• Setting DOCLING_ARTIFACTS_PATH to hardcode RapidOCR downloads by @jamesbraza in #1147
• Fixing jpg handling in ParsedMedia.to_image_url by @jamesbraza in #1150
• Generalizing tests for smarter LLMs by @jamesbraza in #1149
• Supporting page range in readers for performant PDF peeking by @jamesbraza in #1152
• chore(deps): update actions/download-artifact action to v6 by @renovate[bot] in #1155
• Generalizing tests for smarter LLMs 2 by @jamesbraza in #1154
• Deduplicating media on Context creation by @jamesbraza in #1153
• chore(deps): lock file maintenance by @renovate[bot] in #1157
• Recording page_num in media metadata by @jamesbraza in #1158
• Better crash on None value for fields_to_overwrite_from_metadata by @jamesbraza in #1159
• Simplifying prompt newlines to use --- hrule and removed extra hrules by @jamesbraza in #1156
• Supporting media enrichment in embeddings by @jamesbraza in #1143
• Exposed reader_config setting to set DPI and simplify settings by @jamesbraza in #1160
• Removing a few chunk_size/overlap references missed in #1160 by @jamesbraza in #1161
• 'Full page screenshot' enrichment prompt template by @jamesbraza in #1162
• Allowing str year values in paper_search by @jamesbraza...

v5.29.1

What's Changed

• Fixing gen_answer failover leaving raw_answer blank by @jamesbraza in #1077

Note @jamesbraza did a clean rebase of a few commits to beyond this release, so this CHANGELOG entry was manually constructed.

Full Changelog: v5.29.0...v5.29.1