v0.27.2

New features

• Agentic RAG: Refactored the Agentic RAG retrieval framework, significantly improving reasoning speed and benchmark performance. (#19046, #19112, #19172, #19209, #19424, #19459)
• Knowledge compilation runtime settings (#19254)
• Enforce a rate limit on the knowledge compilation pipeline (#19156)
• Knowledge graph shows total and current node counts (#19142,#19095, #19114)
• Graph search highlights the node when pressing Enter on an entity name (#19320)
• New Sitemap data source for sitemap.xml-based web ingestion (#19344)
• WebDAV connector supports custom CA certificates (#16901)
• MonkeyOCRv2 parsing support (#18887)
• Self-hosted PaddleOCR-VL deployment support (#19011)
• New Sofya search tool for agent (#19323)
• EPUB document preview (#19242)
• Excel citation sources located in document preview (#18853)

Model Support

• New Hubris model provider (#19341)
• New llmman model provider (#19030)

Improvements

• New RERANK_TOKEN_LIMIT_MODE configuration (#19404)
• Highlight in retrieval testing and search (#18981)
• Similarity threshold displayed as a percentage (#18974)

UI/UX

• Completed Turkish locale (#19009)
• Search match highlighting in Team settings tables (#15985)

Bug fixes

• Fixed images uploaded in chat being ignored by vision models (#18884)
• Fixed non-UTF-8 encodings in CSV/TXT preview and document decoding (#19222, #19253)

What's Changed

• fix(agentic_rag): include configured system prompt in RAGTools.sys_prompt() by @chakshu-dhannawat in #18927
• fix(web): make sidebar i18n reactive and refine template card/icon UI by @yooxdb in #16646
• fix(dataset): align parser methods order with python by @6ba3i in #18877
• feat(webdav): support custom CA certificates by @taek105 in #16901
• Fix Azure embedding endpoint handling by @zcxGGmu in #16993
• fix go: migrate missing table chatchannel by @myf-bee in #18944
• Python && Go: fix parse status by @JinHai-CN in #18841
• fix(ollama): preserve native API base URL in OllamaCV by @iwasaki1112 in #17013
• fix(file): stop is_parent_folder_exist raising on a missing parent by @marmar9615-cloud in #18930
• fix(agent): enforce MCP task deadlines in worker by @Lem0nTea2002 in #18869
• fix: upgrade starlette to 1.3.1 (CVE-2026-54283) by @anupamme in #18947
• refactor: delete the unused discord_svr.py bot script by @noron12234 in #17682
• Refactor: make all url inside tooltip as link by @wangq8 in #18943
• Py: add update_time in select list when list data sync logs by @JinHai-CN in #18867
• fix: include filename in file download Content-Disposition header by @huang-aoqin in #17105
• fix(embed): drop_params: True was rejected by Together AI and other strict providers by @Harsh23Kashyap in #18627
• fix(rag): key Infinity field_map by original column names, not pinyin by @Taranum01 in #18318
• feat(go-syncer): add Zendesk connector by @mkaaad in #18961
• Fix: Remove the empty response field from the retrieval operator. by @cike8899 in #18963
• fix(agent): reject fractional integer DSL parameters by @6ba3i in #18964
• Fix: update tooltip url to correct link by @wangq8 in #18966
• fix: Introduce argparse to resolve hardcoded parameter values in the SDK test file by @Ricardo-shuo-liu in #16763
• Fix dataset LLM tooltip by @wangq8 in #18971
• fix(web): block saving an agent when an edited Parser node is invalid by @dcc123456 in #18967
• fix(ingestion): carry component outputs in the pipeline operation log DSL by @euvre in #18774
• feat(web): display similarity threshold as a percentage by @dcc123456 in #18974
• fix(agent): let global variable textareas resize vertically by @dcc123456 in #18972
• fix(web): stop the note node resize handle from covering its textarea by @dcc123456 in #18969
• fix(dataflow): decode non-UTF-8 HTML files before parsing in Go pipeline by @euvre in #18801
• fix(parser): drop merged-block and wrapped PDF TOC entries when remove_toc is enabled by @euvre in #18793
• Fix: On the dataset configuration page of the Go version, no error message appears when saving if mandatory fields are left blank. by @cike8899 in #18970
• Fix the tooltip wrong url by @wangq8 in #18975
• fix(admin): stop silent superuser creation on auth probe and harden bootstrap by @simpleqt in #18702
• docs(deepdoc): point the build examples at the Dockerfile's real path by @marmar9615-cloud in #18820
• Fix: For datasets with doc, docx, or markdown formats, the "remove original directory" toggle fails to save successfully. by @cike8899 in #18976
• fix(ingestor,nats): recover consumption, fail fast on NATS startup and harden reconciliation (P0 blockers) by @jay77721 in #18942
• Fix(chat): reorder Rerank and Cross Language form fields in ChatPromptEngine by @cike8899 in #18978
• Fix: remove deprecated qwen tts model by @Lynn-Inf in #18977
• feat(models): add 13 Volcengine Ark chat models by @chyroc in #18973
• fix(agent): make user added emppty entries fail to save by @6ba3i in #18979
• Feature: add highlight for retrieval testing and search by @wangq8 in #18981
• docs：fix some bugs of documents by @Yannnnnnny in #18984
• Fix: pass the kb id to the wiki page insert-error warning by @muhzuhaib in #18985
• Refine agentic search by @yingfeng in #18798
• Refactor agentic search by @yingfeng in #19000
• feat(parser): zero-alloc rune count, bmp/tiff decoders, and clean OCR gate tests (#18794) by @avikumart in #18829
• feat(dataflow): cap debug-run chunk preview and surface run limits in tooltip by @xugangqiang in #18881
• fix(agent): persist webhook traces as JSON objects by @T1mn in #18883
• feat: locate Excel citation sources in document preview by @huang-aoqin in #18853
• fix(agentic_rag): return calculate tool result by @yzl0ng in #19012
• Refactor: update similarity threshold tooltip from 0.2 to 20 by @wangq8 in #19033
• Revert "Refactor: update similarity threshold tooltip from 0.2 to 20" by @wangq8 in #19034
• Refactor: update similarity threshold tooltip from 0.2 to 20 by @wangq8 in #19036
• fix(agentic_rag): normalize structure answers by @yzl0ng in #19025
• Fix: Limit the input box for automatic tags to only accept integers by @cike8899 in #19038
• Refactor agentic search by @yingfeng in #19040
• Fix llm.py by @qinling0210 in #19037
• Feature: support self-hosted PaddleOCR-VL deployments by @maxiaolong03 in #19011
• fix(parser): make PDF "remove TOC" and "remove header/footer" options work without DeepDoc inference by @euvre in #18602
• feat(deepdoc): in-process static-linked ONNX Runtime DeepDoc backend by @xugangqiang in #18731
• feat(go-syncer): add OneDrive connector with lazy fetch and resume by @mkaaad in #18968
• fix(agent): allow deleting edges inside iteration on canvas by @huang-aoqin in #16925
• fix(web): warn about empty Message component content on agent save by @euvre in #18993
• fix(rag/nlp): include question_tks when reranking with a rerank model by @immuhammadfurqan in #18473
• fix(api): return parsing-status counts at the top level of each dataset (#17595) by @Anai-Guo in #17611
• Fix: pass Qwen3 enable_thinking via chat_template_kwargs by @FJH886 in #17592
• fix(common): redact OAuth / API credentials from ConnectorRunner error logs by @Harsh23Kashyap in #17588
• fix(rag): exclude overflow chunk from kb_prompt token budget by @Harsh23Kashyap in #16377
• fix(chat): keep reasoning chats grounded when LLM lacks tool calling by @georgelichen in #18983
• Add go build files , reduce image size, update entrypoint-go.sh by @myf-bee in #19041
• fix(agent): sanitize Excel output sheet names by @uczltw6 in #18505
• fix(migrations): parse the --config flag the usage line documents by @marmar9615-cloud in #18817
• feat(go-syncer): add Airtable connector with lazy fetch and resume by @mkaaad in #19032
• fix(web): default NumberInput precision to 2 by @dcc123456 in #19051
• fix(docker): let the backend launcher retry loops actually run by @marmar9615-cloud in #18926
• fix: allow special tokens in token counting by @buua436 in #19052
• Fix: KnowledgebaseService.dfs_update crashes on stored None parser_config values by @BaptisteJosh in #15964
• docs: add search docs by @Yannnnnnny in #19057
• Fix: pipeline with compilation + transfomer (extrator) does not work by @wangq8 in #19045
• fix(agent): warn at save time when an Agent node has no model selected by @euvre in #18996
• fix(dify): search the dataset owner's index, not the caller's by @marmar9615-cloud in #18987
• fix(rag): return 1-D vector from BaiduYiyan encode_queries by @Harsh23Kashyap in #16398
• feat(web): complete the Turkish locale and fix broken placeholders by @bakiburakogun in #19009
• Fix: Clicking the save button on the Go version of the dataset configuration page automatically locates the field where the error occurred. by @cike8899 in #19061
• fix(sdk): reject conflicting dataset ID filters by @WOLIKIMCHENG in #19047
• fix(memory): restore Python-side rules in Go TYPE_INSTRUCTIONS by @anneheartrecord in #19002
• Restore visible Markdown list styling in chat by @georgelichen in #19010
• fix(serenedb): keep the available_int=0 chunk filter by @marmar9615-cloud in #18928
• fix(confluence): stop passing start/end to _retrieve_all_slim_docs by @Anai-Guo in #18995
• fix: retrieve with the assistant's configured settings by @martincivan in #19016
• feat(go-syncer): add Asana connector with resume and prune by @mkaaad in #19049
• fix(web): shorten the empty-state copy for the wiki tab by @dcc123456 in #19062
• fix(chat): images uploaded in chat UI were ignored by vision models (agentic loop / Go attachment drop) by @euvre in #18884
• fix(web): bound port and max records in the Execute SQL form by @dcc123456 in #19072
• Feat: List all file type parsing configurations for the parser operator. by @cike8899 in #19050
• docs: delete redundant docs by @Yannnnnnny in #19078
• fix: always use full hierarchical path for Confluence attachment/page identifiers by @OSHA-B in #16711
• fix: compare datasets by resolved embedding model name by @yiheng-kkk in #19069
• fix(test): give the rag_agent dialog stub the retrieval settings it now reads by @marmar9615-cloud in #19081
• Go: refactor by @JinHai-CN in #19084
• fix(search): honor the cached document-deleted flag so deleted docs stay pruned by @marmar9615-cloud in #19070
• fix(agent): prevent concurrent webhook trace overwrites by @T1mn in #19026
• feat(agent): expose Iteration max_concurrency on the canvas by @huang-aoqin in #19066
• feat(go-syncer): add SeaFile connector by @mkaaad in #19073
• fix(agent): avoid shared StringTransform merge arguments by @alaningtrump in #19023
• Fix don't send an empty tools array by @martincivan in #19015
• fix(go): parse PPTX one chunk per slide by @euvre in #18254
• fix(ingestion): make NATS task scheduling crash-safe by @jay77721 in #19074
• Refactor agentic search by @yingfeng in #19046
• fix(go-syncer): always use full hierarchical path for Confluence iden… by @mkaaad in #19085
• fix(web): fall back to backend defaults for unset search LLM parameters by @LuckTerence in #19079
• refactor: add context_length for list_models and refactor content_length by @Haruko386 in #19075
• CI : fix ci step ragflow_native_backend requires the following condition by @myf-bee in #19086
• Fix 2 continuous images, the first image got lost by @wangq8 in #19083
• Refactor: add total/returned entities/relations by @wangq8 in #19095
• fix(nats): use DiscardNew for all JetStream work-queue streams by @jay77721 in #19099
• fix(agent): finalize interrupted webhook traces by @T1mn in #19089
• fix(memory): put the keyword weight in the text slot of the fusion expression by @marmar9615-cloud in #18989
• fix(agent): reject only truly-empty delimiter entries on canvas save by @euvre in #19048
• fix(web): honor webhook trace terminal status by @T1mn in #19087
• feat(go-syncer): add BigQuery connector by @mkaaad in #19106
• Fix: The test database connection failed and repeated errors. by @cike8899 in #19102
• Fix: Allow deletion of selected tags in the extractor operator form. by @cike8899 in #19097
• fix(agent): prevent duplicate tools on one Agent node at creation by @euvre in #19053
• Go: add context, part5 by @JinHai-CN in #19108
• fix(document-preview): keep xlsx sheet tabs on one row, drop duplicate fetch by @dcc123456 in #19105
• fix(memory): use the shared NumberInput for the temperature field by @dcc123456 in #19109
• Fix: check minerU.net connection by @Lynn-Inf in #19115
• fix error where returns a no dataset found error in a 500 wrapper by @6ba3i in #19059
• fix(agentic_rag): table-aware retrieval truncation fixes + action-session loop guards by @Caobaining1 in #19112
• Fix: Use default pipeline configuration when parsing documents via HTTP API- #13297 by @JinHai-CN in #19116
• docs: update glossary docs by @Yannnnnnny in #19077
• fix(agent): propagate deferred stream errors through Message by @Lem0nTea2002 in #18796
• build: bump office_oxide to v0.1.9 and pdf_oxide to v0.3.73 by @xugangqiang in #19124
• build(deps): bump nanoid from 3.3.11 to 3.3.18 in /web by @dependabot[bot] in #19128
• fix(deepdoc): harden ORT build guard, OCR test contract by @xugangqiang in #19082
• Fix(parser): raise legacy .doc parsing fidelity via structured IR by @xugangqiang in #18192
• build(deps): bump google.golang.org/grpc from 1.82.1 to 1.83.1 by @dependabot[bot] in #19127
• fix(utils): round minimum_should_match fraction instead of truncating by @RahilOp in #19031
• fix(docker): disable stale macOS nginx mount by @dynnfa in #19126
• fix(test): expect the bare attachment disposition the download path emits by @marmar9615-cloud in #19132
• test(chunker): pin current applyChildrenDelimText behavior (#17876 partial coverage) by @Taranum01 in #18315
• fix(rag/app/table): keep table cells that column type conversion cannot represent by @OmarFaig in #18460
• fix(file): Fix link folder always use 'add' and use API result to refresh linked datasets by @dcc123456 in #19134
• fix(integration): register in-process DeepDoc backend + correct stale chunk/knn assertions by @xugangqiang in #19107
• Refactor: add total/returned entities/relations 2 by @wangq8 in #19114
• fix(tagging): stop a non-object LLM reply from failing the whole batch by @marmar9615-cloud in #18991
• Feat: Displays the total number of nodes and the current number of nodes in ArtifactForceGraph. by @cike8899 in #19142
• fix(agent): restore message download button when output file type is selected by @euvre in #18955
• build(deps): bump nltk from 3.10.0 to 3.10.3 by @dependabot[bot] in #19129
• fix(mcp): bound metadata cache and prevent partial pagination caching by @6ba3i in #19138
• feat(ingestor): raise default max_concurrent_workers from 1 to 2 by @jay77721 in #19133
• fix(ingestion): preserve run input on resume and detach context for terminal cleanup by @jay77721 in #19088
• fix(agent): reject empty message rows when saving canvas by @6ba3i in #19149
• Fix: link to existing dataset clean the document chunks by @wangq8 in #19143
• Fix: hide forgotten message by default. by @Lynn-Inf in #19144
• fix(parser): propagate cancellation while waiting for vision figure futures by @iwasaki1112 in #17014
• test(deepdoc): add 1.pdf colspan end-to-end replay test by @xugangqiang in #19139
• refactor(go-syncer): route GetConnector access errors through helper by @mkaaad in #19146
• Fix NameError in audio chunker when file extension is missing or unsupported by @blueOf in #19007
• chore(deepdoc): source native testdata from infiniflow/ragflow-testdata by @xugangqiang in #19136
• feat(nats): add shared NatsKVStore helper for cache namespaces by @xugangqiang in #19148
• fix: align wiki alteration provenance by @buua436 in #19140
• fix(mcp): read total_datasets so the fetch-all early-exit works by @xu-jia-ming in #19020
• Fix: restrict tenant updates to the current user's own tenant by @Lynn-Inf in #19155
• Refactor: delete not used code by @wangq8 in #19164
• Fix: check required dataset_id/memory_id on save by @Lynn-Inf in #19125
• Fix: Remove the mode from the compiler operator. by @cike8899 in #19150
• fix(dify): return the custom document metadata the handler already fetches by @marmar9615-cloud in #19123
• docs: update documentation to align with current Python API implementation by @Yannnnnnny in #19163
• fix(deepdoc): raise TextDetector limit_side_len default from 960 to 2048 (#18509) by @Harsh23Kashyap in #18888
• refactor(pptx): simplify buildPPTXJSONSections + drop Python-compat wording (Follow-ups 1+2 of #19076) by @Harsh23Kashyap in #19098
• fix(parser): prefer structured IR over flatter views for .doc by @xugangqiang in #19147
• docs(examples): add document lifecycle API examples by @6ba3i in #19153
• Fix: get rid of graphrag raptor in parser_config in REST API by @wangq8 in #19161
• build(deps-dev): bump browserslist from 4.28.1 to 4.28.8 in /web by @dependabot[bot] in #19170
• fix(file): classify ancestry authorization failures as data errors by @6ba3i in #19157
• fix(web): let the editable description textarea scroll when clamped by @dcc123456 in #19167
• Fix: Set the rerank model for the search page as a required field. by @cike8899 in #19154
• Fix: The select-all option in the dataset file list and file management now only applies to the current page; switching pages will clear the previous selection. by @cike8899 in #19158
• Fix: invalidate user when change password or set user as inactive by @wangq8 in #19186
• build(deps): bump pypdf from 6.15.0 to 6.16.1 by @dependabot[bot] in #19171
• Stop splitting resume documents into per-page parsing tasks by @devYRPauli in #18713
• Adjust navigation tools for agentic search by @yingfeng in #19172
• fix(parser): stop the .msg date test depending on the host timezone by @devYRPauli in #19182
• fix(searchbots): enforce KB authorization before retrieval by @6ba3i in #19165
• build(deps): bump qs from 6.14.0 to 6.16.0 in /web by @dependabot[bot] in #19192
• build(deps): bump fast-uri from 3.1.5 to 3.1.7 in /web by @dependabot[bot] in #19191
• Refactor: get rid of graphrag and raptor during task creation by @wangq8 in #19194
• Fix: The backend API did not return the default blueprint template from the wiki. by @cike8899 in #19131
• build(deps): bump mistune from 3.3.0 to 3.3.3 by @dependabot[bot] in #19193
• Require exactly two cells for a CSV QA pair by @devYRPauli in #19008
• fix(py-go-connector): enforce kb authorization in rebuild by @mkaaad in #19190
• feat(ingestion): surface queued document task status by @jay77721 in #19137
• fix(security): generate random UUIDv4 identifiers by @6ba3i in #19159
• refactor(ingestion): unify memory task execution with heartbeat leases by @jay77721 in #19189
• fix(pdf): make _table_transformer_job honor TABLE_AUTO_ROTATE by default by @helloxjade in #16079
• fix(storage): make bucket deletion work on the AWS_S3 backend by @Huang-404-Q in #19162
• Fix OAuth redirect URI examples by @zz80900 in #15683
• docs: add macOS vm.max_map_count setup guidance by @helloxjade in #19177
• Refactor: go default template no more _0 by @wangq8 in #19204
• Fix/final answer streaming phase is not interrupted by the research budget by @Caobaining1 in #19209
• Dev: launch_backend_service.sh to support --debug by @wangq8 in #19215
• Fix: Hide Skills Entry by @cike8899 in #19205
• Fix: Install only oxlint and oxfmt to ensure CI checks for web files. by @cike8899 in #19197
• refactor(parser): collapse docElementText into docxElementText by @xugangqiang in #19135
• fix(storage): handle versioned S3 bucket deletion by @6ba3i in #19199
• fix(rag): unify parser_config delimiter defaults on a single constant by @euvre in #19058
• fix(dataset): fail closed on unknown permissions by @6ba3i in #19213
• [Go] Fix switch embeding report error by @wangq8 in #19201
• fix(dynamic-form): allow non-string values in Segmented fields (#15106) by @dcc123456 in #19212
• fix(storage): scope configured single-bucket S3 removals correctly by @6ba3i in #19217
• feat: add llmman as a model provider by @ericcurtin in #19030
• Fix: The filter data did not update after deleting the agent's item. by @cike8899 in #19220
• fix(preview): handle non-UTF-8 encodings in CSV/TXT preview by @dcc123456 in #19222
• fix: normalize reasoning stream markers by @buua436 in #19216
• refactor deepdoc backend register by @xugangqiang in #19224
• refactor(go): merge messagefit into internal/tokenizer by @jay77721 in #19225
• fix(agent): accept user_id node param on the Retrieval agent tool by @euvre in #19090
• Docs: update indexer by @wangq8 in #19227
• fix(model_meta): discover TEI reranker models by @dawNotPoi in #19210
• fix(agent): wire document service so dataflow rerun works instead of 500ing by @euvre in #19094
• feat: adapt knowledge compilation concurrency by @buua436 in #19156
• fix(ingestion): route xlsx QA JSON table items to extractQATable by @leilei3167 in #19198
• fix ci unit test not found in installed package 'wordnet' by @myf-bee in #19251
• fix(parser): support non-UTF-8 decoding across text, csv, and epub parsers by @jay77721 in #19243
• Fix missing content_ltks when rerank by @wangq8 in #19230
• fix(chunker): align canvas TokenChunker default delimiters with General parser by @xugangqiang in #19226
• chore(deps): depend on org mirror infiniflow/onnxruntime_go by @xugangqiang in #19169
• test(web): cover decodeBlobText across all five decoder branches by @Harsh23Kashyap in #19240
• fix(parser): decode non-UTF-8 document text by @6ba3i in #19253
• fix(token_utils): build the tiktoken encoder on first use, not at import by @marmar9615-cloud in #18986
• Fix: upload image to model in chat by @Lynn-Inf in #19257
• fix(ingestion-logs): normalize dataset status filters by @6ba3i in #19261
• fix(ingestion): show data pipeline on failed parse log entries by @euvre in #19231
• fix(web): sanitize worksheet names that crash the excel previewer by @dcc123456 in #19260
• Fix: check required fields when use template by @Lynn-Inf in #19228
• Fix: After clearing data from pages other than the first page, the data in the filters and search box should also be cleared. by @cike8899 in #19255
• feat(preview): add EPUB document preview with encoding normalization by @dcc123456 in #19242
• feat: configure knowledge compilation runtime settings by @buua436 in #19254
• fix: exclude empty documents from structure alteration by @buua436 in #19246
• fix(web): default Textarea resize to vertical by @dcc123456 in #19263
• Fix: apply dia.llm_id overrides by @Lynn-Inf in #19214
• Fix: bound model response reads by @Lynn-Inf in #19211
• fix(agent): await MCP transport cleanup before stopping event loop by @Lem0nTea2002 in #19093
• fix(dataset): surface Go ingestion QUEUED status and honor terminal run state by @dcc123456 in #19249
• fix(parser): recover invalid XLSX sheet names by @jay77721 in #19264
• CI: add Dockerfile_go build and api test by @myf-bee in #19232
• fix(web): make tr locale capitalisation consistent for expanded labels by @bakiburakogun in #19273
• feat(task_executor): add MAX_TASKS_PER_WORKER to recycle long-running workers by @HamsterPark in #14501
• fix(task): refresh stale embedding model config by @ump45nose in #18089
• fix(parser): handle MinerU chart blocks instead of silently dropping them by @yaoper in #19096
• fix(chunker): prevent chained over-carry of overlap-head PDF positions by @xugangqiang in #19068
• Perf: link onnxruntime without --whole-archive to cut binary size by @xugangqiang in #19277
• docs(api): use the memory update field names the route reads by @marmar9615-cloud in #19279
• fix(api): return the full error tuple when both id and ids are given to list documents by @marmar9615-cloud in #19280
• fix(cli): pass search number as file limit by @helloxjade in #19285
• Fix: Delete icons that have been placed in the assets directory of iconfont.js. by @cike8899 in #19300
• chore: remove the DeepDoc Go alignment report by @xugangqiang in #19311
• fix(web): disallow folder upload in chunk creating modal by @dcc123456 in #19309
• Go: refactor by @JinHai-CN in #19312
• Refactor: update dataset API should not have ext by @wangq8 in #19272
• docs: clarify user_default_llm deprecation by @liwu96 in #19321
• Fix: Click to create an agent; if the name already exists, an error will be displayed in the name input field. by @cike8899 in #19324
• fix(agent): hold back a node whose upstream runs in the same batch by @ebarkhordar in #19282
• fix(memory): restore Infinity FIFO eviction and maintenance queries by @AnnaSuSu in #19290
• fix(storage): honor client region in Python S3 bucket creation by @AnnaSuSu in #19291
• fix(web): load chunk method example images from remote repo by @dcc123456 in #19328
• feat(llm): forward OpenAI user on embedding and parse requests by @ondery in #19065
• fix(mcp): bound the document-metadata cache like the dataset cache by @xu-jia-ming in #19022
• fix(exesql): serialize DATE/TIME columns for canvas SSE JSON by @leilei3167 in #19266
• fix(mcp): terminate Go Streamable HTTP sessions by @Lem0nTea2002 in #19151
• docs: update API references to match implementation by @Yannnnnnny in #19334
• feat(parser): add MonkeyOCRv2-Parsing support by @zenosai in #18887
• fix(web): block chunk save and show error when tag is not selected by @dcc123456 in #19337
• fix: report wiki map timeout errors by @buua436 in #19329
• Fix: After clearing the data in the agent list, the corresponding filter data must also be cleared. by @cike8899 in #19336
• fix(excel): drop the "None" label for a blank spreadsheet header cell by @marmar9615-cloud in #19175
• build: use a stable dynamic-list path for OrtGetApiBase export by @xugangqiang in #19302
• fix(api): allow setting language on dataset creation (#15703) by @manan-tech in #16242
• fix(agent/tools): run a DuckDuckGo news search when the channel is news by @marmar9615-cloud in #19176
• Fix the remaining rag/app chunkers treating an empty upload as a missing binary by @JessYanCoding in #19256
• test(rag/nlp): cover the BOM and LookupError branches of decode_text by @Harsh23Kashyap in #19292
• fix: report knowledge compilation LLM failures by @buua436 in #19313
• Fix: update GiteeAI base_url and some model_name by @Lynn-Inf in #19338
• fix(native): build 0-origin raster in FromImage/Decode for cropped sub-images by @xugangqiang in #19308
• fix: support agent input mapping and converged edges by @buua436 in #19317
• fix(ingestion): keep Q&A/Tag chunks from collapsing into one indexed chunk by @euvre in #18480
• Refactor: combine 3 retrieval/search API as one by @wangq8 in #19303
• fix(api): accept repeated memory_type and tenant_id filters when listing memories by @marmar9615-cloud in #19281
• feat(agent): add Sofya search tool by @yusufgurdogan in #19323
• fix(parser): recover text_box and nested table/list text in office IR by @jay77721 in #19319
• fix(memory): align extracted message document id with message_id by @jay77721 in #19326
• feat(deepdoc): switch in-process Go backend from .onnx to FlatBuffer (.ort) models by @xugangqiang in #19318
• fix: filter structure alteration by document products by @buua436 in #19346
• fix(web): disable textarea resizing in agent note node by @dcc123456 in #19349
• Fix: show skills folder by @wangq8 in #19351
• fix(agent): flatten scalar-list values in DataOperations combine by @WOLIKIMCHENG in #19111
• fix(web): hide sign up until registration is enabled by @shixy96 in #19343
• fix(docs): document dataset language create and update behavior by @6ba3i in #19372
• feat(connector): add Sitemap data source for sitemap.xml-based web ingestion by @vincentlambert in #19344
• fix(llm): raise a clear error for unsupported Bedrock embedding models by @kimnamu in #16044
• feat(channels): allow chat channels to connect to agents by @Gonghan-Princess in #19307
• fix(paper): bound title-pivot merge by chunk_token_num (#12109) by @deadtrickster in #16959
• Dockerfile_go optimize by @myf-bee in #19385
• fix(web): align parse status badge colors with list dots and rename table header by @dcc123456 in #19400
• fix(web): move artifact delete button into the toolbar row by @dcc123456 in #19381
• Fix: By default, the wiki page does not display the wiki content on the right. by @cike8899 in #19388
• fix(agent): clear stale template dataset references by @6ba3i in #19399
• Refactor: Add RERANK_TOKEN_LIMIT_MODE by @wangq8 in #19404
• fix(agent): preserve sandbox artifacts through streaming by @6ba3i in #19355
• fix: resolve Go structure graph template names by @buua436 in #19373
• fix: filter chunks by requested IDs by @buua436 in #19377
• fix(api): return "search not found" instead of IndexError when updating a missing search by @euvre in #19380
• fix(api): enforce readonly access to team-shared chat and agent sessions by @euvre in #19340
• fix(sdk): preserve JSON document downloads by @AnnaSuSu in #19367
• fix(syncer): anchor sitemap PDF-pass checkpoints and declare regex dependency by @mkaaad in #19383
• fix(api): authorize dataset and workspace commit routes by @Harsh23Kashyap in #19357
• fix(agent): treat missing List Operations input as empty list by @euvre in #19275
• fix(dataset): honor keywords filter in Go wiki artifact list APIs by @euvre in #19310
• fix(compilation): make graph search Enter on an entity name highlight the node instead of rendering the whole graph by @euvre in #19320
• feat(ingestion): consolidate checkpoint to Parser + per-chunk result cache by @xugangqiang in #19386
• fix: avoid KeyError when database config has no password (#11051) by @RykerFeng in #16064
• fix(mineru): align backend names with current MinerU API by @huang-aoqin in #19055
• Fix: check sql component required fields by @Lynn-Inf in #19418
• fix(embedding): use Replicate run API for query embeddings by @LostSnowfluff in #19405
• refactor(ingestion): use worker-driven task pull dispatcher by @jay77721 in #19314
• fix(agentic_rag): reject prose fan-outs, keep research loop alive on SCA timeout, and widen slot-table retry budget by @Caobaining1 in #19424
• Fix: The newly added group name for the VariableAggregator operator cannot be found. by @cike8899 in #19440
• fix: prevent IndexError in PUT /api/v1/searches/ when search not found by @MKP999 in #16027
• fix: report incremental wiki map errors by @buua436 in #19422
• chore(deps): let download_go_deps.py fetch Go DeepDoc .ort weights by @xugangqiang in #19376
• fix(compiler): resolve global built-in template scope by @6ba3i in #19426
• Refactor: Follow #19404 to add RERANK_TOKEN_LIMIT_MODE=truncate|passthrough|raise_error by @wangq8 in #19421
• Fix: table column auto parse missing metadata by @wangq8 in #19439
• fix(chunk): enforce document scope on updates by @WOLIKIMCHENG in #19221
• feat: add Hubris as a model provider by @Aimagine-life in #19341
• fix(agent): template retrieval binding at creation + false "no dataset selected" save warning by @euvre in #19316
• feat: highlight search matches in Team settings tables by @carlh7777 in #15985
• Fix: run ensure_db_init before start server by @Lynn-Inf in #19423
• CI: Optimize api test cases failed by @myf-bee in #19450
• Go: update build.sh and development.md by @JinHai-CN in #19448
• Correct parser_config examples in API docs by @AlightSoulmate in #16002
• Doc: remove space by @JinHai-CN in #19462
• fix(llm): implement async chat methods for GoogleChat (Vertex AI Gemini) by @glu000 in #15994
• add pytest-durations to the test workflow by @myf-bee in #19443
• feat(agentic_rag): add tool playbook, clarify tool specs, stop wasted… by @Caobaining1 in #19459
• fix(storage): overwrite existing Azure SAS objects by @AnnaSuSu in #19458
• fix(storage): accept tenant arguments in encrypted Azure calls by @AnnaSuSu in #19457
• Test: Adjust the priority of some test cases from P1/P2 to P3. by @asiroliu in #19461
• build(deps): bump github.com/buger/jsonparser from 1.1.1 to 1.1.2 by @dependabot[bot] in #19463
• fix(restful-tests): align Go parser config contract by @6ba3i in #19442
• fix(agent): preserve Starter values beside query inputs by @6ba3i in #19420
• Decode a single-part email body by its Content-Transfer-Encoding by @devYRPauli in #19178
• fix(embedding): preserve zero filename embedding weight by @LostSnowfluff in #19397
• feat(ingestion): polish image dispatch with bmp/tiff decoders and zero-alloc RuneCount (fix #18794) by @Jaymr1031-Z in #18988
• Fix: NOT generate graph all entities/relations as a graph chunk by @wangq8 in #19474
• Fix: The default name of the operator does not include a numeric suffix. by @cike8899 in #19451
• docs: Update tooltip url by @wangq8 in #19484
• fix(llm): drop Claude sampling params rejected by Bedrock and Anthropic by @vincentlambert in #19295
• Fix: Enabling the opening message switch for the 'begin' operator will cause previously sent messages to be overwritten. by @cike8899 in #19492
• fix(ingestion): allow re-parsing a document whose parse was canceled at progress 0 by @euvre in #19258
• [Refactor] Use regex to do hightlight by @wangq8 in #19488
• fix(go): stop masking document preview failures as "document not found" and add tenant access check by @euvre in #19348
• fix: route navigation API through tree search by @buua436 in #19493
• fix: support keyword search for dataset structure graphs by @buua436 in #19473
• Fix: num bind datasets in template by components by @Lynn-Inf in #19454
• Fix: num token usage by @Lynn-Inf in #19495
• Docs: Update version references to v0.27.2 in READMEs and docs by @asiroliu in #19501
• Docs: v0.27.2 release notes by @wangq8 in #19504

New Contributors

• @chakshu-dhannawat made their first contribution in #18927
• @noron12234 made their first contribution in #17682
• @chyroc made their first contribution in #18973
• @avikumart made their first contribution in #18829
• @maxiaolong03 made their first contribution in #19011
• @Anai-Guo made their first contribution in #17611
• @FJH886 made their first contribution in #17592
• @georgelichen made their first contribution in #18983
• @uczltw6 made their first contribution in #18505
• @BaptisteJosh made their first contribution in #15964
• @martincivan made their first contribution in #19016
• @yiheng-kkk made their first contribution in #19069
• @alaningtrump made their first contribution in #19023
• @LuckTerence made their first contribution in #19079
• @Caobaining1 made their first contribution in #19112
• @dynnfa made their first contribution in #19126
• @OmarFaig made their first contribution in #18460
• @blueOf made their first contribution in #19007
• @xu-jia-ming made their first contribution in #19020
• @Huang-404-Q made their first contribution in #19162
• @zz80900 made their first contribution in #15683
• @ericcurtin made their first contribution in #19030
• @dawNotPoi made their first contribution in #19210
• @leilei3167 made their first contribution in #19198
• @HamsterPark made their first contribution in #14501
• @liwu96 made their first contribution in #19321
• @zenosai made their first contribution in #18887
• @JessYanCoding made their first contribution in #19256
• @yusufgurdogan made their first contribution in #19323
• @shixy96 made their first contribution in #19343
• @kimnamu made their first contribution in #16044
• @Gonghan-Princess made their first contribution in #19307
• @RykerFeng made their first contribution in #16064
• @LostSnowfluff made their first contribution in #19405
• @MKP999 made their first contribution in #16027
• @Aimagine-life made their first contribution in #19341
• @AlightSoulmate made their first contribution in #16002
• @Jaymr1031-Z made their first contribution in #18988

Full Changelog: v0.27.1...v0.27.2