v0.27.1

New features

• New Azure DevOps connector for data sources (#18715)
• New You.com web search provider for chat and agent (#18478)
• New Serply web search provider for chat (#18475)

Model Support

• New Synthorai model provider (#18830)
• Missing DeepSeek models added (#18678)
• AWS Bedrock API key authentication support (#18301)

Improvements

• Retrieval API exposes rerank_candidates_count, knn top_k, and num_candidates (#18768, #18737)
• Metadata filters pushed down to the metadata index for faster retrieval (#18219)
• Chat settings form auto-scrolls to the error location on save (#18811)
• Search page validates deleted knowledge bases (#18522)

Bug fixes

• Fixed PDF parsing failure in chat (#18860)
• Fixed wiki compilation accepting LLM error responses (#18857)
• Fixed MinerU rendering stability, error visibility, and code block parsing (#18617)
• Fixed 0-chunk datasets not showing, preventing scrolling to fetch more datasets (#18810)
• Fixed QA Excel parser dropping rows whose answer is zero (#17902)
• Fixed naive parser treating an empty upload as a missing binary (#18826)
• Fixed encoding detection discarding confident detection results (#15793)
• Fixed reasoning-enabled chat ignoring the dialog system prompt (#18842)
• Fixed knowledge compilation using the default tenant LLM instead of the pipeline LLM (#18572)
• Fixed chat network search service not allowed to be null (#18791)
• Fixed Firefox chat auto-scroll not pinned to the bottom (#18736)
• Fixed chat auto-scroll not following streamed answers (#18570)
• Fixed parsing progress not visible after selecting upload (#18741)
• Fixed BigQuery connector constructing SQL queries incorrectly (#17500)
• Fixed WebDAV listing errors not propagated, causing prune data loss (#18694)
• Fixed cross-KB document ID collision in data source sync (#18244)
• Fixed GPUStack model discovery and audio support (#18599)
• Fixed Kimi K3 temperature (#18564)
• Fixed OCR text lost when no image2text model is configured (#18012)
• Fixed hierarchical Wiki topic navigation (#18721)

What's Changed

• fix(web): style think/retrieving blocks in floating chat widget by @dcc123456 in #18512
• fix[syncer]: github data source don't validate repo's accessible by @Haruko386 in #18477
• fix: test data source connection need connector name by @Haruko386 in #18495
• feat[syncer]: add data source Confluence by @Haruko386 in #18501
• refactor: remove some unused model types by @Haruko386 in #18502
• fix(api): report why provider model discovery failed instead of swallowing it by @olifarhaan in #18027
• fix: preserve extensionless document suffix on GaussDB by @Sevenzuo in #18483
• fix(memory): honor custom extraction prompts by @yzl0ng in #18461
• fix(memory): normalize invalid extraction timestamps by @yzl0ng in #18463
• fix(memory): require semantic valid_at timestamp by @yzl0ng in #18462
• fix(nlp): differentiate alphabetic OOV term weights by @yzl0ng in #18470
• fix: chat did not pass model config to provider by @Haruko386 in #18485
• feat(chat): self-check contradictory prompt/dataset settings (#5703) by @peewee92 in #18274
• feat(chat): add Serply web search provider by @googio in #18475
• fix: clear stale tenant model ids by @01JohnMa in #18210
• Feat: connector go imap by @mkaaad in #18513
• fix: pass Infinity vector similarity weight by @Hellrakk in #17453
• TS: update thinking description by @JinHai-CN in #18521
• fix: keep OCR text when no image2text model is configured by @Dhruv-kys in #18012
• fix(chat): drain deep-research goroutines before closing the async channel by @euvre in #18504
• feat(deepdoc): Python-intermediate replay parity harness + table cell-fill fixes by @xugangqiang in #18516
• fix(ollama): preserve multimodal images by @lntutor in #17663
• feat(cli): add --version flag to cli by @krishna231122-del in #18534
• Add ulimits for cpu and gpu services in Docker Compose by @Ro0tk1t in #17205
• fix: ollama chat error by @Haruko386 in #18540
• fix: avoid quadratic dedup in RAG merge paths by @DeviousDrops in #18135
• refactor(chunker): replace QAChunker/PresentationChunker with PairChunker/PageChunker, drop TagChunker by @xugangqiang in #18523
• Update release notes by @JinHai-CN in #18544
• feat(chat): add You.com web search provider by @brainsparker in #18478
• Fix: respect IME composition state on Enter key in dataset creating dialog by @iwasaki1112 in #17150
• fix(dataset): preserve disabled RAPTOR and GraphRAG settings by @lntutor in #17665
• Fix: handle tenant_llm without id by @Lynn-Inf in #18541
• refactor(ingestion): unify auto-metadata on modular metadata and hard-delete legacy flats (BuiltInMetadata carried, not LLM) by @jay77721 in #18511
• feat(title-chunker): enforce chunk_token_cap on the Go title-family chunks (#18455) by @xugangqiang in #18551
• fix(deepdoc): drop nested OCR boxes before table cell fill + expose 13/14 parity gaps by @xugangqiang in #18524
• fix: data source return two error once it occured by @Haruko386 in #18535
• feat[syncer]: implement dingtalk AI table data source by @Haruko386 in #18539
• docs: add dataset docs & fix broken links in docs by @Yannnnnnny in #18553
• test: pin bare tenant_model.id resolution in resolve_model_config by @pangwangshu in #18528
• fix(memory): make default prompts deterministic by @primorLee in #18547
• fix(auth): strip trailing slash from OIDC issuer before discovery by @krim404 in #18530
• feat(syncer): add R2 connector with checkpoint resume by @mkaaad in #18548
• feat(chunker): enforce a hard token cap in the TokenChunker merge by @xugangqiang in #18525
• fix: get two VLM model type when list all models by @Haruko386 in #18546
• fix(db): make negative GET_LOCK timeouts work on MariaDB by @krim404 in #18529
• fix: preserve blocking GaussDB advisory locks by @Sevenzuo in #18506
• feat: improve incremental Wiki compilation by @buua436 in #18557
• Fix: ensure tenant_ocr_id is added by @Lynn-Inf in #18560
• Fix: switch image/ppt/email/audio from General to pipeline, or switch back, it report error by @wangq8 in #18555
• fix(compilation-template): emit wiki presets with the contracted "example" key by @euvre in #18549
• Fix: go switch user from superuser to normal user error by @wangq8 in #18537
• fix(ingestion): show data pipeline column for builtin pipeline ingestion logs by @euvre in #18484
• feat(syncer): add Moodle connector with checkpoint resume by @mkaaad in #18554
• refactor: simplify sync task service and harden scheduler recovery by @Haruko386 in #18550
• feat[syncer]: implement data source Jira by @Haruko386 in #18558
• fix: render both initials for multi-word names in RAGFlowAvatar by @tryeverything24 in #17320
• feat(agent): support OpenAI-compatible responses by @T1mn in #18316
• fix: align wiki graph and dataset logs by @buua436 in #18566
• fix: apply order_by return value in deep-pagination helpers (unordered OFFSET/LIMIT can skip/dupe rows) by @rsnetworkinginc in #17325
• fix(web): sanitize the parsed markdown tree instead of the string (GHSA-jv5f-3mqh-cxwf) by @kah-ja in #17590
• fix(deepdoc): dedup OCR substring fragments regardless of whitespace divergence by @xugangqiang in #18568
• fix(deepdoc): preserve covered cells through span injection so span tables keep all TSR rows by @xugangqiang in #18574
• Go: fix error handle by @JinHai-CN in #18573
• fix: store source key only in pipeline_operation_log.source_from by @euvre in #18239
• fix(parser): allow text output_format for audio family by @xugangqiang in #18575
• fix(agent): reset execution path for new sessions by @yzl0ng in #18532
• Feat: Validate the knowledge base that was deleted from the search page. by @cike8899 in #18522
• fix(web): keep chat auto-scroll following streamed answers by @dcc123456 in #18570
• Fix knowledge compilation to use pipeline LLM instead of default tenant LLM by @wangq8 in #18572
• Fix: kimi k3 temperature by @Lynn-Inf in #18564
• fix(web): preserve entire subtree of title-matching nodes in TreeSelect filter by @dcc123456 in #18561
• fix: implement Tongyi-Qianwen TTS via DashScope native endpoint by @euvre in #17924
• fix(agent): bind query field in agentbot completions and clean up failed first-touch sessions by @euvre in #17844
• fix(file): reject folder names containing "/" in create/rename by @euvre in #18586
• fix(memory): persist memory valid_at/forget_at as server-local time instead of UTC by @euvre in #18538
• Fix: propagate file rename to all linked knowledgebase documents by @euvre in #18590
• fix(models): infer model types on catalog miss so Qwen VLMs show LLM+VLM tags by @euvre in #18571
• Fix: Go chat pipeline sends image blocks to text-only models (Zhipu GLM error 1210) by @euvre in #18580
• fix(engine): return error instead of panic when NATS engine is uninitialized by @euvre in #18536
• Feat: Rewrite the Go-based extractor operator form. by @cike8899 in #18600
• Fix: allow yiyan api key by @Lynn-Inf in #18601
• fix(doc_store): retry transient connection errors in pool init by @Harsh23Kashyap in #17196
• fix(search): fix layout clipping when search page is not fullscreen by @euvre in #18519
• fix(search): do not jump to previous search results on empty query by @euvre in #18520
• feat(go-syncer): add Slack connector with checkpoint resume by @mkaaad in #18603
• perf(retrieval): push metadata filters down to the metadata index by @ting-hong-shieh in #18219
• add dockerfile_ci for workflow and restore the single dockerfile by @myf-bee in #18606
• docs: add links between docs by @Yannnnnnny in #18609
• fix(web): clicking anywhere in a switch setting row toggles the switch by @euvre in #18567
• fix(metadata): batch large document ID filters by @guominjia in #17394
• fix: dataset ingestion log has no other type instead of success by @Haruko386 in #18605
• fix(llm): consolidate TTS / Seq2txt provider key resolution across 3 call sites by @Harsh23Kashyap in #17688
• fix: preserve empty chat fields on GaussDB by @Sevenzuo in #18445
• feat[provider]: apply model_type infer when model has no type by @Haruko386 in #18593
• feat[syncer]: add data source DropBox by @Haruko386 in #18612
• optimize[syncer]: optimize returning error when it occur by @Haruko386 in #18613
• fix(build): name the compiler the check requires, and the option that failed by @marmar9615-cloud in #18594
• fix: preserve non-string values in update_metadata_to by @huang-aoqin in #17346
• Feat/connector go sharepoint by @mkaaad in #18618
• Refactor: Refactored the frontend logic for handling different backend services to use pickByBackend, thereby improving code clarity and maintainability. by @cike8899 in #18614
• fix(web): keep multi-select count in sync after deleting a single file by @euvre in #18589
• Fix: Owner filter count not updated after deleting a dataset by @euvre in #18623
• Fix(go): use llm setting for AI summary and return raw error msg by @Lynn-Inf in #18615
• fix(web): accept a single API key for Baidu YiYan by @dcc123456 in #18611
• fix(deepdoc): close ocr_real text-gap divergences and pin char-to-box matching by @xugangqiang in #18630
• fix(cli): count benchmark application failures by @NgoQuocViet2001 in #18292
• fix(show_env): keep git errors out of the environment report by @marmar9615-cloud in #18597
• docs(entrypoint): show every option usage() accepts by @marmar9615-cloud in #18596
• fix(deepdoc): align Go table row count with Python per-char R/C in parity replay by @xugangqiang in #18622
• fix(deepdoc): resolve OCR text gaps — char loss/duplication in pdf parser by @xugangqiang in #18643
• Fix HTML parser treating an empty upload as a missing binary by @devYRPauli in #17904
• port smart-reasoning agentic RAG to eino ADK by @yuzhichang in #18654
• fix(web): give each provider instance card its own surface by @dcc123456 in #18620
• fix(ingestion): honor auto-keyword/auto-question top_n sliders in extractor prompts by @euvre in #18632
• fix(web): resize the mind map graph when its container changes size by @dcc123456 in #18616
• fix(web): use the real Keenable mark for the tool icon by @ilya-bogin-keenable in #18660
• add nats and clickhouse expose port in docker compose by @myf-bee in #18577
• fix: Invoke proxy uses validated URL; remove stray debug print in Canvas.reset by @simpleqt in #18657
• Remove three no-op self-assignments by @darkdi in #18652
• docs: replace root-relative cross links with relative page links by @simpleqt in #18664
• fix(opensearch): stop zeroing vector scores by mapping similarity to knn boost by @barandras in #18662
• Revert "port smart-reasoning agentic RAG to eino ADK" by @JinHai-CN in #18670
• Fix docs format by @yingfeng in #18675
• fix: account for checkpointed ingestion progress by @buua436 in #18667
• fix(workerpool): close the SubmitTo/StopWait race without losing in-flight tasks by @changshenhan in #18646
• fix(manual): match .docx only so legacy .doc fails cleanly by @ayam04 in #18649
• feat(go-syncer): add Salesforce connector with checkpoint resume by @mkaaad in #18674
• fix(ingestion): refactor tag extractor with memory index, asymmetric coverage, multi-source enrichment and worker pool sequencing by @jay77721 in #18607
• security: remove leaked OAuth credentials and stop logging user secrets by @simpleqt in #18658
• Fix: correct rerank #17296 - add prefetch_size by @wangq8 in #18300
• feat(go-syncer): add Teams connector with positional resume by @mkaaad in #18680
• fix: precompile regexes on the retrieval hot path by @changshenhan in #18644
• Fix: porting #18300 to go by @wangq8 in #18690
• fix(web): hide empty thinking strip in floating chat widget by @euvre in #18608
• fix(agent): preserve whitespace around message references by @primorLee in #18666
• fix: validate message content to prevent empty values by @6ba3i in #18676
• fix: the pdf generation issue in agent by @6ba3i in #18673
• Chore: add missing deepseek model by @6ba3i in #18678
• fix(sdk): temperature default should be a float, not a 1-tuple by @simpleqt in #18656
• fix(task_executor): cap extracted important_kwd terms to ES keyword limit by @RahilOp in #17512
• Fix some comments by @wangq8 in #18705
• fix: normalize model types in _factory_model_types and update calculate_model_type by @6ba3i in #18685
• fix: recover panics in executeMemoryTask so a poison memory task cannot kill the ingestor worker by @xugangqiang in #18687
• feat(extractor): add inline 24h Redis LLM cache for keywords, questions, and summary by @jay77721 in #18716
• fix(deepdoc): prevent cross-page table over-merge from page-local Y coordinates by @xugangqiang in #18688
• Feat: The extractor operator can handle multiple extraction tasks simultaneously. by @cike8899 in #18717
• fix(chat): stop rag_agent forwarding message bookkeeping keys to the provider by @bharadwaj-pendyala in #18663
• fix(deepdoc): propagate spanning-cell bbox so Go emits correct colspan by @xugangqiang in #18707
• fix: publish wiki event when document is deleted by @buua436 in #18682
• fix(agent): surface cancellation terminal state by @T1mn in #18579
• fix(models): pass selected models into verify and catch constructor errors by @ondery in #17132
• feat(bedrock): support API key authentication by @whhe in #18301
• fix: classify .wmf files as VISUAL instead of OTHER by @devYRPauli in #15822
• Refactor: rename prefetch_size to rerank_candidates by @wangq8 in #18718
• fix(wiki): guard generation without compilation config by @libaojiang in #18695
• docs(migration): say which volumes the backup actually covers by @marmar9615-cloud in #18626
• fix(api): make the API-key subject carve-out on POST /messages reachable by @olifarhaan in #18026
• feat(agent): add You.com search tool by @brainsparker in #18583
• fix(rag): keep search queries and page text out of connector logs by @brainsparker in #18582
• fix(admin): display AWS S3 status correctly by @yaoper in #17300
• [Fix] Refined the NLTK data usage directory by @paulhuiseismic in #17376
• fix: substitute UserFillUp tips placeholders literally to avoid re.error by @rsnetworkinginc in #17321
• fix: substitute Message variables literally to avoid re.error on regex escapes by @rsnetworkinginc in #17319
• fix: remove trailing space in human-readable duration labels by @tryeverything24 in #17317
• fix(web): return real navigation state from history.location getter by @tryeverything24 in #17324
• fix(web): ignore IME composition Enter in tag input by @tryeverything24 in #17322
• Fix: show linked knowledge base names instead of IDs when searching in the "Link to knowledge base" dialog by @euvre in #18581
• fix(web): show agent opening statement in embedded floating widget by @euvre in #18517
• fix(bot): align chatbot completion final frame and history persistence with python by @euvre in #18526
• Fix: Add a custom placeholder to SelectWithSearch by @cike8899 in #18719
• Fix: add and use paddleOCR.net model by @Lynn-Inf in #18696
• fix(web): keep multi-select options in their declared order while searching by @dcc123456 in #18672
• fix(web): make chat attachment images click-to-zoomable by @euvre in #18259
• fix(web): make overflowing card description tooltips scrollable by @euvre in #18700
• fix(setting-model): stop silently dropping a duplicate-named provider instance by @dcc123456 in #18722
• fix(go): sort provider model instances by creation time by @euvre in #18684
• fix(web): chat settings panel overlapped by chat input on narrow viewports by @euvre in #18723
• Fix: After deleting all compiled entities, returning to this page will show an empty data entry. by @cike8899 in #18724
• fix(web): always show copy button on floating chat widget replies by @euvre in #18527
• fix(router): accept share (beta) token on agent upload and attachment download by @euvre in #18681
• fix(svr): distinguish empty-docs from missing-template in wiki empty-eligibility message by @Harsh23Kashyap in #18691
• fix(common): propagate WebDAV listing errors from _list_files_recursive to avoid prune data loss by @Harsh23Kashyap in #18694
• fix(agent): bound crawler response bodies to 4 MiB by @Lem0nTea2002 in #18677
• fix: support hierarchical Wiki topic navigation by @buua436 in #18721
• Fix email and presentation parsers treating an empty upload as a missing binary by @devYRPauli in #18712
• Go: porting #18663 by @wangq8 in #18726
• fix: skip empty knowledge compile events by @buua436 in #18728
• Fix: Modify the error message for empty chat responses. by @cike8899 in #18730
• fix(mcp): include dataset name in list_datasets output (#17133) by @Harsh23Kashyap in #18710
• fix(connector): honor configured S3-compatible bucket type by @yijiaquan in #18576
• fix(pdf): colspan assembly loss + TSR geometry comparison diagnostics by @xugangqiang in #18727
• Fix: cross-KB document id collision in data source sync (#18116) by @sarbojitrana in #18244
• Refactor: expose knn top_k and knn num_candidaties by @wangq8 in #18737
• feat(go-syncer): add Azure Blob Storage connector with checkpoint resume by @mkaaad in #18701
• fix(admin): honor DEFAULT_SUPERUSER_EMAIL/PASSWORD env vars in bootstrap by @Taranum01 in #17674
• Fix: After selecting upload, the document is parsed directly, but the parsing progress cannot be seen. by @cike8899 in #18741
• fix(web): fall back a page when the dataset file list empties by @dcc123456 in #18738
• fix(web): require chunk content when creating or editing a chunk by @dcc123456 in #18739
• Fix: Add the chat model to the Python version of the dataset configuration page. by @cike8899 in #18735
• fix(web): keep Firefox chat auto-scroll pinned to the bottom by @dcc123456 in #18736
• fix(agent): prevent path traversal and SSRF in Browser URL file upload by @simpleqt in #18711
• fix(agent): redact internal terminal errors by @T1mn in #18743
• Fix GPUStack model discovery and audio support by @javen-yan in #18599
• Porting #18737 to go by @wangq8 in #18745
• fix(pdf): restore cross-page table merge wrongly rejected by #18688 page-absolute Y gate by @xugangqiang in #18740
• Fix total as the total, follow Go by @wangq8 in #18750
• fix(agent): panic and raw error when auto play runs without a default TTS model by @euvre in #18734
• fix(chat): propagate mind map stream failures by @Lem0nTea2002 in #18733
• fix(parser): crop media sections in PDF JSON path, unify Data URI and markdown rendering by @jay77721 in #18742
• fix(deepdoc): harden opendataloader table-cell parsing against bad json by @Yurii214 in #17316
• fix(pdf): strip tiled watermark text from text-layer extraction by @Taranum01 in #18308
• fix(agent): default missing Graph path so stored DSL cannot KeyError by @BetterAndBetterII in #18753
• Fix(go): PaddleOCR -> PaddleOCR.local, PaddleOCR.net -> PaddleOCR by @Lynn-Inf in #18747
• fix(chat): deliver uploaded image attachments to vision models in Go chat pipeline by @euvre in #18708
• feat(data_source): add Azure DevOps connector by @erhanaydin in #18715
• fix(memory): align Go semantic valid_at prompt with the Python wording by @anneheartrecord in #18752
• docs(sandbox): describe how the container pool is actually sized by @marmar9615-cloud in #18625
• fix(sandbox): remove the sandbox containers that exist, not a pool-size range by @marmar9615-cloud in #18629
• fix(web): surface dataset navigation tree load failures (closes #17301) by @Souravrajvi0 in #17668
• Docs: update 27.0 release notes by @wangq8 in #18770
• fix(parser): clean .eml text output for chunking by @euvre in #18592
• fix(web): agent explore streaming answer leaking into the switched-to session by @euvre in #18686
• fix(web): stop RAGFlow brand block from overlapping embedded share header by @euvre in #18729
• fix(sandbox): sync the sandbox env with the Python version it requires by @marmar9615-cloud in #18766
• docs(api): spell the recommendation endpoint the way the server registers it by @marmar9615-cloud in #18764
• docs(api): document the document-update endpoint as PATCH by @marmar9615-cloud in #18765
• Fix: MinerU model by @Lynn-Inf in #18767
• Fix: The recommended chunk size for the token chunker is restricted to integers only. by @cike8899 in #18786
• fix(agent): propagate webhook terminal states by @T1mn in #18758
• fix: allow empty GitHub repositories by @buua436 in #18776
• Fix: find_codec discards confident encoding detection and returns a garbage codec by @devYRPauli in #15793
• fix: inherit document status when inserting chunks by @huang-aoqin in #18610
• fix(naive): route misconfigured MinerU PDFs to by_mineru instead of by_plaintext (#17114) by @Taranum01 in #18762
• refactor(parser): consolidate vision figure enhancement across document formats by @jay77721 in #18775
• fix(parser): support webp in image OCR and correct rune count for VLM gate by @jay77721 in #18785
• fix: the bigquery connector constructs sql queries u... in... by @anupamme in #17500
• fix(opensearch): scope multi-document update to the target knowledge base by @AnnaSuSu in #17048
• fix: accept empty ASR verification results by @buua436 in #18788
• fix(web): cap embed dialog User ID input length and add data-type hint by @euvre in #18773
• fix(model-provider): resolve instances by persisted ID by @6ba3i in #18792
• fix(docker): prevent all foreground entrypoint.sh loops from exiting on non-zero exit by @skbs-eng in #18545
• feat(data-source): add Xquik X search connector by @kriptoburak in #18714
• fix(web): skip canvas autosave of empty nodes/edges by @Souravrajvi0 in #18783
• fix(sdk): filter documents by multiple IDs by @WOLIKIMCHENG in #18790
• fix(web): use a dedicated label for the thinking mode field by @dcc123456 in #18806
• Fix: The chat network search service is allowed to be null. by @cike8899 in #18791
• fix(security): authenticate sandbox executor API and isolate container network by @simpleqt in #18703
• refactor(parser): extract shared parseWithTCADP core from pptx/xls/pdf wrappers by @Taranum01 in #17671
• Fix 0 chunk datasets not show, cause cannot scroll to fetch more datasets by @wangq8 in #18810
• fix(chat): preserve deep-research cancellation after workers finish by @Lem0nTea2002 in #18692
• fix(mineru): Fix MinerU Rendering Stability, Error Visibility, and Code Block Parsing Issues by @Aeovy in #18617
• fix(web): list the agent flow type card first by @dcc123456 in #18808
• Chore: remove knowledge compiler from types and tests by @6ba3i in #18809
• Feat: When there are errors in the chat settings form, clicking the "Save" button automatically scrolls to the location of the error. by @cike8899 in #18811
• Refactor: Add rerank_candidates_count, knn top_k and num_candidates to /api/v1/retrieval by @wangq8 in #18768
• fix(go): drop the removed internal/cache package from run_go_tests.sh by @marmar9615-cloud in #18819
• fix(build): keep the admin server in build.sh's cleanup trap by @marmar9615-cloud in #18816
• fix(tests): let run_tests.py --help exit 0 by @marmar9615-cloud in #18814
• Go: fix context, part3 by @JinHai-CN in #18821
• feat: extract images from xlsx parser by @buua436 in #18800
• fix(pdf): do not cross-page merge tables whose anchor is not at page bottom by @xugangqiang in #18789
• fix(document): reject config edits while document is parsing or scheduled by @xugangqiang in #18804
• fix(parser): align image output_format to json and use per-family defaults (strict) by @jay77721 in #18802
• Better enhance message by @wangq8 in #18818
• fix(graph): use max-wins for PageRank in AnalyzeNHopPaths + correct test fixtures (#15695) by @Taranum01 in #18784
• fix: align cross-language prompt separator with parser by @zhouyuchong in #18831
• docs(configuration): name the ragflow serving-port key http_port by @marmar9615-cloud in #18825
• Fix: An error occurs when accessing the Go version of the wiki page. by @cike8899 in #18835
• Refactor: refine error messages - use %W by @wangq8 in #18834
• fix(pdf): remove unused context import breaking manual test build by @xugangqiang in #18837
• fix(schema-sync): stop reporting BaseModel timestamps as dropped columns by @marmar9615-cloud in #18824
• fix(chunker): keep PDF positions on all cap-split sub-chunks so chunks don't lose their preview images by @euvre in #18772
• fix(examples): wait for the DONE status the API actually returns by @marmar9615-cloud in #18827
• Fix the naive parser treating an empty upload as a missing binary by @devYRPauli in #18826
• Refine entrypoint.sh - tidy code by @wangq8 in #18838
• Fix: On the MCP list page, if the only record on the last page is deleted, the view should navigate to the previous page. by @cike8899 in #18840
• Feat: sync anchor resume by @mkaaad in #18795
• Fix QA Excel parser dropping rows whose answer is a zero value by @devYRPauli in #17902
• Update None to Naive (Follow English) by @wangq8 in #18848
• feat: add Synthorai model provider by @cuihuan in #18830
• Fix OceanBase chunk order field persistence by @zcxGGmu in #16997
• fix: fall back to the real office container format when extension is wrong by @jay77721 in #18843
• Refactor: Translate figure 'Fig.' to Chinese by @wangq8 in #18852
• fix(api): honor dialog system prompt in reasoning-enabled chat by @RahilOp in #18842
• feat(agent): make MCP tool call timeout configurable by @var4dev in #18016
• fix(agent): implement webhook trace polling by @T1mn in #18815
• fix(web): keep typed slider values on the slider's step grid by @dcc123456 in #18847
• fix(dataset): canonicalize naive parser IDs to general by @6ba3i in #18839
• fix(web): cap description length at 512 characters by @dcc123456 in #18836
• Fix: The models for audio and video parsers are required fields. by @cike8899 in #18851
• Fix: remove deprecated mimo-v2-flash by @Lynn-Inf in #18854
• fix(web): add missing hover tooltips to dataset data source action buttons by @euvre in #18812
• feat(go-syncer): add OCI Storage connector with resume by @mkaaad in #18850
• test(agent): pin whitespace handling of template ref patterns by @ayam04 in #18720
• fix(web): dataset auto-metadata toggle reverts to off after save by @euvre in #18813
• Fix: When the pipeline-paeser component switches the file type from PDF to image, the default parsing method should be OCR. by @cike8899 in #18858
• fix(web): keep token chunker settings when saving in One mode by @euvre in #18749
• fix(nlp): load term weight resources as UTF-8 by @yzl0ng in #18467
• fix: support .wmf file extension case-insensitively and remove duplicate avi by @Wsp030914 in #15498
• fix(parser): wire per-call VLM model config, harden vision dispatch by @jay77721 in #18846
• Revert "fix(web): file list total count doesn't match displayed rows (#17747)" by @dcc123456 in #18856
• fix(ingestion): run wiki compiler cache-less in canvas debug runs by @euvre in #18828
• fix: reject wiki LLM error responses by @buua436 in #18857
• Fix: parse pdf in chat by @Lynn-Inf in #18860
• Fix: hide skills folder in REST API by @wangq8 in #18859
• Fix rerank_candidates_count validation error by @wangq8 in #18863
• fix(metadata-filter): propagate resolved document IDs to retrieval by @6ba3i in #18865
• Fix Go Agents tab ordering to match Python by @6ba3i in #18845
• feat(go-syncer): add S3 and S3 compatible connectors by @mkaaad in #18861
• fix(chunker): proportionally slice PDF positions so sub-chunk previews match text by @jay77721 in #18864
• Fix: No need to validate data that is not in dataset format by @cike8899 in #18871
• Fix: rm default model in create and display builtin model by @Lynn-Inf in #18875
• docs: add Chat documents by @Yannnnnnny in #18870
• fix(web): derive provider base_url visibility from the provider catalog by @dcc123456 in #18866
• fix: handle Jina VLM streaming responses by @buua436 in #18855
• fix: map agent rounds to react steps by @buua436 in #18868
• Fix oversized important keywords in OceanBase and SeekDB by @hnwyllmm in #18872
• fix(rag/utils): ensure_v1 must recognize v1beta/v1alpha-style version suffixes by @chuenchen309 in #16916
• fix: honor generation parameter enable flags by @buua436 in #18880
• fix: support SeekDB vector score extraction by @kic635 in #18874
• Refactor: add page_size support by @wangq8 in #18876
• Fix: add Separator component and restore GlobalIndexModelItem in NaiveConfiguration by @cike8899 in #18882
• fix(api): derive SQL retrieval tenant from dataset owner, not chat tenant by @ChendongPang in #18518
• fix(web): stop the field label from opening the MultiSelect popover by @dcc123456 in #18922
• Fix: reorder components in NaiveConfiguration for better structure by @cike8899 in #18924
• Fix: Switching the file type of the parser operator caused the corresponding parse method to fail to switch. by @cike8899 in #18931
• Fix: respect IME composition state on Enter key send by @iwasaki1112 in #16922
• fix(build): pass GOPROXY to the ragflow-cli go build by @marmar9615-cloud in #18925
• feat(go-connector): add Box connector by @mkaaad in #18934
• fix(web): stop the field label from toggling the Switch by @dcc123456 in #18945
• fix(dataset): rebuild the navigation tree in the dataset owner index by @marmar9615-cloud in #18932
• Docs: Update version references to v0.27.1 in READMEs and docs by @asiroliu in #18849
• Docs: 0.27.1 release notes by @wangq8 in #18879

New Contributors

• @olifarhaan made their first contribution in #18027
• @yzl0ng made their first contribution in #18461
• @googio made their first contribution in #18475
• @Hellrakk made their first contribution in #17453
• @Dhruv-kys made their first contribution in #18012
• @krishna231122-del made their first contribution in #18534
• @Ro0tk1t made their first contribution in #17205
• @DeviousDrops made their first contribution in #18135
• @brainsparker made their first contribution in #18478
• @primorLee made their first contribution in #18547
• @krim404 made their first contribution in #18530
• @T1mn made their first contribution in #18316
• @rsnetworkinginc made their first contribution in #17325
• @kah-ja made their first contribution in #17590
• @marmar9615-cloud made their first contribution in #18594
• @NgoQuocViet2001 made their first contribution in #18292
• @simpleqt made their first contribution in #18657
• @darkdi made their first contribution in #18652
• @barandras made their first contribution in #18662
• @changshenhan made their first contribution in #18646
• @ayam04 made their first contribution in #18649
• @bharadwaj-pendyala made their first contribution in #18663
• @libaojiang made their first contribution in #18695
• @yijiaquan made their first contribution in #18576
• @javen-yan made their first contribution in #18599
• @BetterAndBetterII made their first contribution in #18753
• @erhanaydin made their first contribution in #18715
• @Souravrajvi0 made their first contribution in #17668
• @kriptoburak made their first contribution in #18714
• @zhouyuchong made their first contribution in #18831
• @cuihuan made their first contribution in #18830
• @Wsp030914 made their first contribution in #15498
• @kic635 made their first contribution in #18874
• @ChendongPang made their first contribution in #18518

Full Changelog: v0.27.0...v0.27.1