This Sycamore release contains a variety of bug fixes and improvements.

What's Changed

• Low-hanging mypy fruit (Sycamore edition). by @alexaryn in #1265
• Miscellaneous code-level improvements. by @alexaryn in #1266
• Improve OpenSearch reader by batching certain operations by @austin-aryn-ai in #1267
• Don't include original_elements in queries by @baitsguy in #1268
• Add a new OpenAI model, revert LlmFilterPrompt change by @austin-aryn-ai in #1271
• Add metadata loading to materialize by @HenryL27 in #1270
• Update httpcore and h11 for dependabot issue. by @bsowell in #1274
• Put os_client_args in default namespace to avoid surprise. by @alexaryn in #1275
• Add debugging for unexpected gemini model stop reasons. by @eric-anderson in #1276
• Summarize the group name for groupby by @bohou-aryn in #1261
• Update torch to 2.7.0 and transformers to 4.50.0. by @bsowell in #1279
• Adding gpt-4.1 and gpt-4.1-nano by @Soeb-aryn in #1278
• Add better view_pdf function by @HenryL27 in #1277
• Standardize materialize filenames a little by @HenryL27 in #1283
• Add support for non-clustering groupby by @bohou-aryn in #1281
• Add instructions to split entities in LLM extract entity. by @akarshgupta7 in #1286
• Add function to unroll entities by @bohou-aryn in #1288
• fix hybrid table model fallback to not edit tokens in-place by @HenryL27 in #1290
• Add heuristics to compute K for kmeans based on docset size. by @akarshgupta7 in #1287
• fix the non cluster groupby path error by @bohou-aryn in #1291
• Speed up import sycamore by ~10x; add module/tool for timing imports. by @eric-anderson in #1292
• Improve docset documentation to explain that it's lazy. by @eric-anderson in #1294
• handle non clustering aggregate count by @bohou-aryn in #1293
• Minor fixups from testing with managed-service. by @eric-anderson in #1296
• load tiktoken tokenizer lazily by @HenryL27 in #1298
• refactor planner to add query preprocessor functionality by @HenryL27 in #1297
• Add Gemini 2.5 Flash by @karanataryn in #1295
• Upgrade gemini by @bsowell in #1299
• Adding notebooks that walk through the earnings calls documents by @AbhijitP-009 in #1284
• Bump AIOHttp To Fix Package Install by @karanataryn in #1300
• OpenSearchReader result filtering by @austin-aryn-ai in #1282
• Add support for default llm kwargs for all models. by @bsowell in #1303
• Use llm for clustering by @bohou-aryn in #1302
• Upgrade tornado dependency. by @bsowell in #1305
• Add VLMTableStructureExtractor for table structure extraction. by @bsowell in #1304
• Fix collect when groupby has no entity names by @bohou-aryn in #1307
• Upgrade setuptools. by @bsowell in #1309
• Add type casting logic to extract entity. by @akarshgupta7 in #1310
• Remove aryn-sdk from the sycamore repo. by @bsowell in #1306
• Initial draft of X-Y Cut reading order for Sycamore (Hackathon) by @alexaryn in #1301
• improve model selection error message by @HenryL27 in #1311
• add id field param to queryresult.retrieved_docs by @HenryL27 in #1312
• Update LLM models by @bsowell in #1313
• Fix the example llm_cluster_instruction for the topk operator. by @vikram-ak in #1314
• Add a split_pdf function. by @bsowell in #1315
• fix case where tokens are None inside of model selection parsing by @HenryL27 in #1317
• Fix fast notebook tests by @bsowell in #1316
• Handle case where query is a compound query by @austin-aryn-ai in #1319
• Add DeepWiki Badge by @karanataryn in #1321
• Fix issue with nested_lookup utility. by @bsowell in #1320
• Remove index from Jinjaprompt to avoid token blow up. by @akarshgupta7 in #1322
• Add flag to skip table extraction on empty tables by @MarkLindblad in #1323
• Enable pdfminer vertical text grouping. by @bsowell in #1324
• Update Slack invite link in README.md by @sohamkasar19 in #1325
• Fix when unroll handles None entity field by @bohou-aryn in #1326
• Add some useful planner processors by @HenryL27 in #1327
• Update Slack invite link in documentation and README.md by @sohamkasar19 in #1329
• Add code to turn materialize off for all nodes after and incl Sort. by @akarshgupta7 in #1328
• Add support for a custom supplement_text function in the partitioner. by @bsowell in #1332
• allow planner customization via prompt specification by @HenryL27 in #1330
• Upgrade to torch 2.7.1. by @bsowell in #1334
• Add RECOMPUTE source mode for nodes with materialized disabled. by @akarshgupta7 in #1333
• Rename Table Extractor Options to Table Extraction Options by @karanataryn in #1338
• Add Ability to Resolve Boundary Overlaps in Tables by @karanataryn in #1337
• Fix materialize commit by @bohou-aryn in #1331
• Add OpenAI reasoning models. by @bsowell in #1340
• Add retries for VLMTableStructureExtractor. by @bsowell in #1341
• Fix Resolve Overlaps Plumbing by @karanataryn in #1339
• First part of reliable opensearch writing -- handles new items and missing metadata on source or destination. by @eric-anderson in #1335
• Add support for document reconstruct for RAG. by @akarshgupta7 in #1343
• Sycamore: deal with rotated tables. by @alexaryn in #1336
• Add RAG support in Luna. by @akarshgupta7 in #1344
• Fix test clustering flaky by @bohou-aryn in #1345
• Add empty list in RAG doc reconstructor if doc not found in unique_docs. by @akarshgupta7 in #1346
• Fix notebook tests. by @eric-anderson in #1347
• Upgrade dependency on requests by @bsowell in #1348
• Remove default parameter for Gemini max_output_tokens. by @bsowell in #1349
• Introduce a new chained LLM by @austin-aryn-ai in #1342
• Fix MRR refresh by @dhruvkaliraman7 in #1350
• Add logging on retries. by @eric-anderson in #1318
• [tmp!!!] switch off partitioner its by @HenryL27 in #1351
• Load models in eval mode, no gradient computation by @dhruvkaliraman7 in #1352
• Update protobuf version for Dependabot. by @bsowell in #1353
• Add support for new Gemini models. by @bsowell in #1355
• Improve ChainedLLM to handle when to move to next llm in chain, add r… by @austin-aryn-ai in #1354
• Fix more Dependencies by @karanataryn in #1356
• add kwargs to planning processors and plumb through sq client by @HenryL27 in #1360
• Refactor Process Batch by @karanataryn in #1357
• Add Extract Images Function by @karanataryn in #1361
• switch partitioner its back on by @HenryL27 in #1366
• Add source fields to RAG document reconstructor. by @akarshgupta7 in #1364
• Upgrade PaddleOCR by @karanataryn in #1365
• Add size to opensearch knn query. by @akarshgupta7 in #1368
• Fix OCR Bug by @karanataryn in #1369
• Drop support for Python 3.9 and add Python 3.13. by @bsowell in #1367
• [Dependency] Upgrade urllib3. by @bsowell in #1370
• [Dependency] Upgrade urllib3 in one more app. by @bsowell in #1371
• fix sycamore query codegen mode options by @HenryL27 in #1372
• adding Metadata extraction logic to extract_properties by @Soeb-aryn in #1362
• Add Optional Parameters by @karanataryn in #1374
• Add opensearch scores to element properties. by @akarshgupta7 in #1375
• Change the name of search relevance score. by @akarshgupta7 in #1376
• Add default to relevance scores returned by opensearch. by @akarshgupta7 in #1377
• Add support for extract_image_format when using remote ArynPartitioner by @MarkLindblad in #1379
• Support Delete and Update in OpenSearchSync by @eric-anderson in #1373
• Bump Pillow by @karanataryn in #1378
• Bump Pillow by @karanataryn in #1382
• Bump Paddle to 3.1 by @karanataryn in #1383
• Add prompt tail to llm planner by @dhruvkaliraman7 in #1385
• plumb kwargs through to run plan as well as generate plan by @HenryL27 in #1386
• Make OS optional in LLMPlanner by @dhruvkaliraman7 in #1387
• Upgrade transformers to 4.53.1. by @bsowell in #1388
• Fix warning for K-Means by @bohou-aryn in #1390
• add limit llm operations and require query database processors by @HenryL27 in #1389
• Upgrade aiohttp dependency by @bsowell in #1392
• Tolerate mis-formatted files instead of cleaning up. by @eric-anderson in #1380
• Scripts for running docparse controlled by BigQuery by @eric-anderson in #1393
• Change ArynWriter to use new serialization approach by @MarkLindblad in #1384
• Fix ArynWriter's use of stream by @MarkLindblad in #1394
• Upgrade jupyter-core. by @bsowell in #1398
• Bump sycamore version to 0.1.33 by @MarkLindblad in #1397
• Upgrade Gemini 2.5 Flash Lite, which is now GA. by @bsowell in #1399

New Contributors

• @vikram-ak made their first contribution in #1314

Full Changelog: v0.1.32...v0.1.33