infiniflow/ragflow v0.27.3

on GitHub

What's Changed

• Port agentic RAG to GO by @qinling0210 in #19500
• Revert "fix(ingestion): allow re-parsing a document whose parse was canceled at progress 0" by @wangq8 in #19503
• fix(agent/tool): translate MCP tool inputSchema.properties in Info() by @mkaaad in #19244
• Fix: remove deprecated functions by @Lynn-Inf in #19449
• fix(sdk): preserve agent answer references by @AnnaSuSu in #19365
• fix(mcp): send rerank_candidates_count so retrieval paging works past 64 results by @xu-jia-ming in #19018
• fix(agent): compare retrieval KBs by resolved embedding model, not raw embd_id by @kabishou11 in #19206
• Go: add integration test by @JinHai-CN in #19485
• Fix: Modify the text color of the TreeSelect component. by @cike8899 in #19507
• Upon entering a wiki, the top-right corner indicates that modifications have been made, even though no changes were actually performed. by @cike8899 in #19482
• fix(api): authorize task cancellation by dataset access by @Harsh23Kashyap in #19359
• refa: remove unused knowledge compilation code by @buua436 in #19437
• fix: avoid nested compiler pool embedding deadlock by @buua436 in #19486
• fix(invoke): tolerate invalid headers across runtimes by @6ba3i in #19481
• fix(wiki): treat MAP-extracted chunks as cache hits instead of missing by @kabishou11 in #19208
• fix(web): prevent DOCX preview infinite update loop by @Souravrajvi0 in #19475
• fix(web): add Traditional Chinese child delimiter translations by @huyua9 in #18951
• fix(figure_parser): accept lang as named parameter on VisionFigureParser (#17280) by @Taranum01 in #19511
• fix(dataflow): preserve scalar text output shape by @huyua9 in #18921
• fix(i18n): localize Chinese delimiter preview by @huyua9 in #18948
• Unify ONNX Runtime native to 1.29.0 from infiniflow/ragflow-build by @xugangqiang in #19446
• fix(document): validate immutable fields on update by @taek105 in #19444
• fix(structure-graph): discover buckets from raw rows by @6ba3i in #19523
• fix: forward NumberInput ref to input by @huyua9 in #18904
• fix(ssrf): unwrap IPv6 transition addresses in the URL guard by @aroh3006 in #19517
• Feat: Make the files of the parser operators in the pipeline deletable. by @cike8899 in #19384
• fix(template): align data analysis CodeExec prompt by @huyua9 in #18958
• fix(storage): retry transient S3 failures by @6ba3i in #19526
• Port agentic RAG PRs to GO by @qinling0210 in #19527
• refa: remove Go artifact deletion APIs by @buua436 in #19519
• test(infinity): cover SQL literal escaping by @huyua9 in #18956
• Fix: The text in the SelectWithSearch dropdown menu is obscured by the checkmark icon. by @cike8899 in #19534
• fix(build): anchor pdf_oxide version gate against suffix versions by @xugangqiang in #19152
• feat(nlp): rerank on natural chunk text, not tokenized content_ltks by @vincentlambert in #16961
• fix(chunker): one BPE pass per hardSplitPiece iteration instead of two by @changshenhan in #18648
• fix: update knowledge artifacts on document status changes by @buua436 in #19532
• Migrate cache backend from Valkey 8 to apache/kvrocks:2.16.0 by @xugangqiang in #19374
• docs： update FAQs by @Yannnnnnny in #19538
• refactor(parser): unify parser outputs to structured JSON items by @jay77721 in #19467
• feat(agent): support document_ids in retrieval node and tool by @huang-aoqin in #16938
• fix(ingestion): stop dropping QA pairs from multiline table cells by @jay77721 in #19536
• Add claim/evidence to tree/pageindex compilation pipeline by @yingfeng in #19267
• fix(parser): handle large sparse spreadsheets, JSON BOM, and XLSX QA chunking by @Souravrajvi0 in #19237
• fix(wiki): version history always empty on Go API due to page-slug key mismatch by @euvre in #19305
• Port Agentic RAG to Golang by @yingfeng in #19563
• feat(data-source): add Feishu Wiki file connector by @Gonghan-Princess in #19425
• docs: fix dead link to removed docker-compose-gpu.yml by @jrlprost in #19576
• fix(ollama): disable reasoning for vision descriptions by @6ba3i in #19529
• fix(beartype): remove package-wide import hooks by @6ba3i in #19531
• Custom the file extension when saving DB fields, and Parse HTML fragments not just full HTML doc by @wadefelix in #15998
• fix(api): authorize the dataset document download route by tenant in the Go API by @coderdailyone in #19543
• fix: handle indirect PDF annotation arrays by @huyua9 in #18894
• fix(sdk): retain image IDs on chunks by @AnnaSuSu in #19505
• Fix: Set the minimum value for "top n" to 1 and restrict it to an integer. by @cike8899 in #19547
• Fix: Validate webhook form by @cike8899 in #19546
• feat(syncer): add Feishu Wiki connector for the Go backend by @mkaaad in #19453
• feat: add AnonRouter model provider by @anonrouterai in #19561
• fix: surface MinerU API error bodies and re-raise parser failures by @huang-aoqin in #17308
• fix(model): delete provider models and instances in one transaction by @ming1523 in #19562
• fix: fix security issue in ob_conn_base.py by @anupamme in #19524
• feat(connector): support file_extension for MySQL/PostgreSQL sync by @mkaaad in #19466
• feat(channels): WeCom can bind agent by @brucehu in #18598
• fix(wiki): stamp retrieval fields so compiled pages are searchable by @Souravrajvi0 in #18787
• feat(api): add image_update_mode for chunk image updates by @Souravrajvi0 in #18889
• Fix: keep a reference to the qqbot dispatch task so replies are not dropped by @muhzuhaib in #19029
• fix(agent): make Python webhook trace appends atomic by @T1mn in #19091
• fix(rag): assign PDF parser tables to tbls in book chunking method (#19557) by @Ritiky23 in #19559
• fix(storage):preserve original errors after retries by @6ba3i in #19580
• fix(web): scope pipeline log detail query keys by @huyua9 in #18909
• fix(api): validate embedding check sample count by @Harsh23Kashyap in #19574
• fix(api): treat an empty metadata_condition.conditions list as no filter by @Harsh23Kashyap in #19368
• fix(api): intersect metadata filter results with the caller's doc scope by @Harsh23Kashyap in #19353
• fix CI native backend test failed by @myf-bee in #19590
• Remove internal cli 'insert chunks from file' and 'insert metadata from file' by @qinling0210 in #19591
• fix(storage):preserve errors during backend retries by @6ba3i in #19578
• fix: merge same-name knowledge graph entities by @buua436 in #19525
• fix: split knowledge compilation logs by task type by @buua436 in #19548
• fix(parser): emit xlsx spreadsheets as typed table chunks by @xugangqiang in #19509
• Fix: UI chat should pass pass_all_history_messages=false in 1.0 go by @cike8899 in #19602
• Fix: The selected invalid dataset cannot be deleted. by @cike8899 in #19603
• fix: update image utils test stub by @buua436 in #19605
• docs: fix Python frontend proxy startup by @liang0417 in #19606
• fix: align mindmap relations with Python by @buua436 in #19579
• Revert "Fix: UI chat should pass pass_all_history_messages=false in 1.0 go" by @wangq8 in #19618
• refactor(ingestion): decouple document parsing state from document.run in favor of ingestion_task by @jay77721 in #19533
• fix(web): adapt document parsing state for Go ingestion_status (depends on #19533) by @dcc123456 in #19587
• Feat: Hide the delete and update buttons related to the Go version of the wiki. by @cike8899 in #19620
• Fix agentic search bugs by @yingfeng in #19615
• fix(agent): Agent reply empty when a retrieval tool runs — need2cite flips mid-stream by @spicyfishPro in #19483
• fix(agent): preserve Boolean values in Variable Assigner by @LostSnowfluff in #19499
• fix(sdk): stop parsing waits on status lookup failures by @AnnaSuSu in #19506
• feat: add API-Route as a model provider by @DennyHo0917 in #19566
• PY: refactor by @JinHai-CN in #19599
• fix(agent/tools): encode GitHub and QWeather query parameters by @MarMarLabs in #19584
• docs: clarify parser and email text by @gluedtea in #19541
• fix: show document graph entity counts by @buua436 in #19617
• Feat: Shorten the default distance of the timeline connector lines. by @cike8899 in #19612
• Fix: display modelExtra in api response by @Lynn-Inf in #19607
• feat(memory): make Go memory extraction tasks durable and resumable by @taek105 in #19512
• fix(ci): read DeepDoc testdata from the pre-seeded dir instead of linking it in by @xugangqiang in #19594
• test: cover TokenizerFromUpstream empty payload contracts by @huyua9 in #18898
• fix(retrieval): recover dense candidates after empty hybrid search by @6ba3i in #19502
• Implement endpoint GET /datasets/tags/aggregation in GO by @qinling0210 in #19626
• docs: repair stale internal links and license reference by @Mnilax in #19633
• fix(db): abort poisoned transactions on connection loss by @6ba3i in #19530
• Fix: When calling the children and graph interfaces on the wiki tree page, the keywords parameter should be included. by @cike8899 in #19627
• Fix chat bot reasoning in GO by @qinling0210 in #19641
• fix(agent): reject unsupported image input cleanly by @6ba3i in #19539
• feat(mineru): observability for embedded image loss in chunk stream (#16978) by @Taranum01 in #18761
• fix(chunker): align custom delimiter text metadata by @yzl0ng in #19005
• fix(ocr): prevent OpenCV remap crashes with oversized images by clamp… by @lucasoffchain in #19419
• fix(api): reject empty provider instance deletes by @huyua9 in #18919
• fix(ingestion): enforce text-only pre-index wire contract by @Souravrajvi0 in #19478
• fix(api): preserve parent-child config on partial updates by @huyua9 in #18949
• feat(ingestion): track a run's pipeline log from queue to terminal by @jay77721 in #19494
• fix(naive): do not route vision layout models to MinerU by @huang-aoqin in #19513
• ci: drop the kvrocks-parity workflow and the integration tests it ran by @xugangqiang in #19583
• fix: record generated wiki page versions by @buua436 in #19636
• fix: support keywords in navigation APIs by @buua436 in #19645
• fix: sync navigation tree with document status by @buua436 in #19653
• build(deps): bump flask from 3.1.2 to 3.1.3 by @dependabot[bot] in #19659
• build(deps): bump protobufjs from 7.6.4 to 7.6.6 in /api/channels/whatsapp/gateway-node by @dependabot[bot] in #19655
• build(deps): bump mermaid from 11.16.0 to 11.16.1 in /web by @dependabot[bot] in #19657
• build(deps): bump langgraph-sdk from 0.3.6 to 0.3.15 by @dependabot[bot] in #19660
• Fix: The 'image' parsing method has been configured, but uploading images still results in an error. by @cike8899 in #19667
• fix(engine): filter chunks by requested IDs in update-by-query by @qinling0210 in #19665
• refactor(ingestion): migrate general parser and chunker contract by @jay77721 in #19611
• Refactor: move chat message/reference to dedicated tables by @wangq8 in #19577
• fix(web): reactively render children delimiter field by @huyua9 in #18892
• fix: correct hyphens utility on agent template descriptions by @huyua9 in #18901
• fix(parser): attach positions on TCADP spreadsheet JSON items by @li872 in #19555
• fix: resolve template names in keyword graph search by @buua436 in #19640
• fix: filter compilation status by kind by @buua436 in #19666
• Fix the pipeline email parser failing on every multipart message by @JessYanCoding in #19406
• fix(nlp): detect capitalized table-of-contents headings by @chuenchen309 in #17012
• PY: apply cross_languages translation in Retrieval memory mode by @JinHai-CN in #19669
• Fix saving agent without compiler id should report error by @wangq8 in #19672
• feat(web): explain chunker delimiter semantics in the pipeline forms by @jay77721 in #19676
• Go: invite member by send mail by @JinHai-CN in #19675
• Fix(go): migration cleanup and add DB version tracking by @Lynn-Inf in #19598
• Fix: ignore empty dataset config when creating from a template by @Lynn-Inf in #19677
• fix(memories): derive filter options from page list query to prevent stale filters by @dcc123456 in #19537
• fix(web): add table styling in chunk card content by @dcc123456 in #19614
• feat(web): add virtual scrolling to CSV viewer and chunk list by @dcc123456 in #19621
• feat(web): disable LaTeX math rendering for user messages by @dcc123456 in #19624
• feat: add DaoXE as an OpenAI-compatible model provider by @seven7763 in #19560
• fix(agent): expose token and round controls by @6ba3i in #19637
• fix(ragcon): scope embedding drop_params to official endpoint by @6ba3i in #19476
• fix(agent): wire MCP tools and preserve execution errors by @6ba3i in #19628
• Fix: Agent: Clicking multiple times to view the version history and then clicking the initial version causes an error. by @cike8899 in #19674
• Fix: Remove redundant parameters from the Go-version knowledge base configuration page. by @cike8899 in #19668
• feat(ui): add locked option support to MultiSelect component by @dcc123456 in #19658
• Fix: Knowledge Compilation - Graph - Entity count remains unchanged after search. by @cike8899 in #19652
• fix(web): fall back to dataset ID for tag KB IDs on Go backend by @dcc123456 in #19648
• Fix: When the agent compiler adds a new chunk node via the "Next Step" action, it can generate multiple subsequent nodes. by @cike8899 in #19644
• fix(agent): honor multi-select canvas category filter in Go /agents listing by @euvre in #19625
• fix(llm): skip dashscope/ prefix for Tongyi-Qianwen's OpenAI-compatible endpoint by @Taranum01 in #19288
• fix(auth): separate OAuth JIT provisioning policy by @6ba3i in #19651
• fix(web): guard tenant info in embed app modal by @huyua9 in #18940
• feat(web/switch): allow dot-notation field paths in Switch condition (#14235) by @Khaostica in #15071
• Fix: restrict instance name to digits, underscores, and letters by @Lynn-Inf in #19680
• fix(data_source): support custom base URL and HTTP for Azure DevOps c… by @ori-frenkel in #19407
• fix(bedrock): report the expected key schema instead of a JSONDecodeError (#17373) by @Anai-Guo in #17374
• Fix: append attachment content to last user prompt by @wangq8 in #19691
• fix(llm): honor the batch size BaiduYiyanEmbed already declares by @L4XB in #19690
• Fix: use component name instead of ID by @wangq8 in #19700
• Dev: add LLM_DEBUG to debug LLM request and response. by @wangq8 in #19703
• fix(mineru): upload the PDF once when several page ranges are configured by @devYRPauli in #19183
• fix(parser): flush pending text before table/image chunks in _build_cks by @wang393 in #19521
• feat(ingestion): stop persisting business data in pipeline operation-log DSL by @xugangqiang in #19631
• fix(ingestion): include the image/table context window in the chunk text by @jay77721 in #19678
• Fix: allow use - in instance_name by @Lynn-Inf in #19709
• fix(dao): allowlist orderby columns on the search, chat, file and dataset lists by @MarMarLabs in #19601
• fix(api): stop trusting X-Forwarded-For from every peer in the agent webhook IP whitelist by @coderdailyone in #19550
• fix: improve knowledge compiler LLM handling by @buua436 in #19701
• fix(deepdoc): preserve textless PDF figures for VLM extraction (#19679) by @Ritiky23 in #19688
• Fix: Request artifact logs based on the selected kind. by @cike8899 in #19716
• Fix {date} not take effect in system prompt by @wangq8 in #19713
• fix(chunk): gate claims/evidence panels by compilation template kind by @dcc123456 in #19704
• Fix: Dataset file upload: For unsupported file types, prompt the user to configure the corresponding file parser. by @cike8899 in #19710
• fix(pipeline-log): record the pipeline Parser's parse_method, not document.parser_id (#18306) by @Taranum01 in #18763
• Fix: Self-referencing data returned by the API causes the mind map page to freeze. by @cike8899 in #19720
• fix(csv-preview): move border to inner wrapper so it wraps content only by @dcc123456 in #19715
• Fix: enable admin user by default by @wangq8 in #19718
• fix(deepdoc): keep every inference call within one process CPU budget by @jay77721 in #19695
• Fix the One and Book chunkers when Tika extracts nothing from a .doc by @JessYanCoding in #19692
• perf(chunker): make hardSplitPiece O(L) with a single-encode token walk by @changshenhan in #19712
• fix(parser): keep the word boundary a styled space carries into markdown by @L4XB in #19693
• fix(web): correct metadata filter button semantics by @huyua9 in #18911
• fix(ingestion): count child tokens and cut media context on sentences by @jay77721 in #19696
• Fix: keep file API responsive during large connector ingest by @ondery in #19632
• fix(parser): keep a one-row slide table instead of dropping it by @L4XB in #19705
• fix(parser): accept a one-sheet dict in the Excel pandas fallback by @L4XB in #19706
• fix(api): return document metadata from /dify/retrieval by @ebarkhordar in #19686
• Fix user detail show not work by @wangq8 in #19719
• fix(agent): accept string variable selectors in VariableAggregator by @Harsh23Kashyap in #19413
• fix: filter self-loop knowledge compiler relations by @buua436 in #19721
• docs: update knowledge compilation docs by @Yannnnnnny in #19734
• docs: update MCP documents by @Yannnnnnny in #19730
• fix(i18n): rename 'Max tokens' to 'Max context length' and drop misleading default by @dcc123456 in #19729
• Fix: Switching the document-level graph selector should also update the corresponding total node count. by @cike8899 in #19732
• Fix: An error occurs when clicking the virtual root node of the mind map to call the API. by @cike8899 in #19731
• fix(file-commit): stop delete operations crashing the whole commit by @MarMarLabs in #19654
• Optimizing recall for agentic RAG by @yingfeng in #19687
• Feat: migrate chat message and reference by @Lynn-Inf in #19724
• fix(api): align Go proxy contract parity by @6ba3i in #19723
• fix(go): harden connector SSRF by @mkaaad in #19694
• fix(api): clamp memory list tenant_id filter to joined tenants by @coderdailyone in #19739
• fix(syncer): keep the word boundary a styled space carries into markdown by @mkaaad in #19722
• feat(mcp): add SDK-backed native Go MCP runtime by @6ba3i in #19673
• Fix searchbot return 500, should be 200 with error message by @wangq8 in #19763
• fix(agent): parse composite llm_id right-anchored on provider by @coderdailyone in #19752
• feat(deepdoc): verify PaddleOCR service connectivity in check_installation by @HUAN2022A in #17669
• fix(agent): reject missing models before save and run by @6ba3i in #19761
• fix(parser): skip an unreadable EPUB chapter instead of losing the book by @L4XB in #19708
• docs: run migrations before starting Go servers by @QAQadws in #19776
• Feat: If a compilation type is not selected, the toggle for re-splitting Parser output is not displayed. by @cike8899 in #19771
• fix(ingestion): make running transition atomic across created and scheduled by @jay77721 in #19764
• refa: keep builtin compilation templates out of database by @buua436 in #19760
• fix(infinity): repair chunk ingestion and dataset save by @qinling0210 in #19773
• Fix chunks search with keywords not working by @wangq8 in #19772
• fix: change citation labels from "Fig. 1" to "[1]" by @Yannnnnnny in #19781
• Fix tree navigation search with keywords not working by @wangq8 in #19774
• Refactor thinking log of agentic log by @qinling0210 in #19707
• fix(parser): skip table-of-contents entries when parsing docx by @wang393 in #19491
• fix(parser): preserve unwrapped Markdown code fences by @AnnaSuSu in #19552
• fix: align graph keyword search with Python by @buua436 in #19778
• fix(benchmark): account for failed requests and incomplete streams by @ning-jl in #19498
• feat(go-syncer): retry sync task errors with per-class budgets by @mkaaad in #19790
• docs: update data source documentation by @Yannnnnnny in #19798
• Fix: route API-Route to it's driver by @Lynn-Inf in #19793
• feat(agentic_rag): slot-table harness, table rendering + pre-search/metadata channels by @Caobaining1 in #19780
• feat(sandbox): complete Go sandbox settings and live refresh by @6ba3i in #19783
• fix(admin): correct service details lookup by @6ba3i in #19770
• fix(csv-preview): stable horizontal scroll for ragged CSV rows by @dcc123456 in #19785
• Fix: Update the prompt text displayed when clicking to parse text for an unconfigured audio model. by @cike8899 in #19788
• feat(chunk): add min/max validation for tag frequency field by @dcc123456 in #19794
• Refactor: Replace RAGFlowSelect with SelectWithSearch. by @cike8899 in #19795
• feat(datasource): add BackButton to data source detail page header by @dcc123456 in #19796
• Fix: Selecting an option from the Wiki Graph dropdown menu also triggers an API request to fetch the latest data. by @cike8899 in #19801
• fix(web): open variable dropdown from prompt toolbar icon without prior focus by @euvre in #19728
• Feature: go highlight follow python, follow #19488 by @wangq8 in #19792
• Fix citation of agentic rag by @qinling0210 in #19791
• Fix metadata filter issue by @wangq8 in #19803
• fix(deepdoc): disable ORT BFC memory arena for pooled inference sessions by @xugangqiang in #19800
• feat(go-syncer): add checkpoint resume for MySQL/PostgreSQL connectors by @mkaaad in #19762
• fix(api): return empty documents when doc_ids filter is empty by @JinHai-CN in #19809
• Refact agentic search by @yingfeng in #19811
• feat: sort sync log list by status priority by @mkaaad in #19777
• Enhance LLM_DEBUG with took and first token time, short embedding to 3 float by @wangq8 in #19816
• feat: add Cheaper Inference as an OpenAI-compatible model provider by @aiapienthusiast in #19748
• fix(parser): move PDF TOC and header/footer removal to box level by @jay77721 in #19779
• fix(document): skip active and completed tasks on batch ingest by @jay77721 in #19786
• fix(compiler): keep LLM runtime params when saving a document pipeline config by @euvre in #19554
• Fix authorization for OpenAI-compatible agent completions by @MQ-xz in #19817
• Fix: Negative values ​​can be entered for the maximum number of reflection rounds. by @cike8899 in #19819
• For no auth provider, list models will return empty by @JinHai-CN in #19827
• Refine citation for agentic RAG by @yingfeng in #19822
• Fix: Restrict the parser operator so that it can only be followed by a chunker, and a chunker cannot be followed by another chunker. by @cike8899 in #19833
• refactor(agent-form): remove maxTokens switch field from sub-agent form by @dcc123456 in #19830
• fix: resolve provider aliases for extractor models by @buua436 in #19824
• feat(ingestion): add run-scoped progress events by @jay77721 in #19775
• fix(parser): recover DOCX text boxes in the Python chunkers by @L4XB in #19735
• Go: add generic Go integration test CI job (infinity) by @xugangqiang in #19702
• refa: migrate model resolution to ModelSolver by @buua436 in #19820
• fix(model): move PaddleOCR api_key unwrapping from the generic provider flow into the drivers by @euvre in #19823
• fix: improve UI wording and localization for parsing and chunking settings by @Yannnnnnny in #19839
• Fix: store content to conversation_message by @wangq8 in #19836
• Revert "feat(agent): support document_ids in retrieval node and tool" by @wangq8 in #19837
• fix(embedding): omit unsupported user field for OVH AI endpoint by @techpreneur-a in #19758
• fix(sdk): preserve completion message IDs by @AnnaSuSu in #19551
• fix(tts): send required voice payload for SiliconFlow TTS and surface synthesis failures by @euvre in #19812
• fix(document): hide a disabled document's knowledge-compiled chunks from retrieval by @qinling0210 in #19845
• fix(go): surface the underlying error instead of generic 500 for related search and mindmap by @QAQadws in #19832
• fix(ingestion): unify run duration and timestamp parsing log lines by @jay77721 in #19849
• feat(go,web): port tokenizer counters, embedding limits and multi-provider web search by @yuzhichang in #19802
• Fix: The Go version of the artifact tree also needs to display build logs. by @cike8899 in #19841
• fix(admin): show host and port in service status by @6ba3i in #19848
• refactor(go): default Go cache/queue backend to Kvrocks and fail fast when down by @xugangqiang in #19804
• fix: update words in UI by @Yannnnnnny in #19861
• fix(memory): expose server-side filter discovery by @6ba3i in #19840
• fix(agent): report missing Message variables clearly by @6ba3i in #19831
• fix(admin): use GetKvrocksConfig instead of undefined GetRedisConfig by @xugangqiang in #19867
• feat(models): add OpenAI GPT-5.6 model family by @Souravrajvi0 in #19856
• Go: fix MySQL connection port exhaustion under high concurrency by @JinHai-CN in #19881
• Go: add pprof, use --profile to turn on by @JinHai-CN in #19858
• Go: reuse a global HTTP client and tune connection pool to prevent port exhaustion by @JinHai-CN in #19888
• fix(markdown): support non-numeric citation IDs in reference markers by @dcc123456 in #19846
• fix(dataset): skip clear-chunks confirmation when re-parsing on Go by @dcc123456 in #19851
• Feat/xiaomi list models refresh by @QAQadws in #19855
• fix(agent): keep deferred Agent model failures in the chat stream instead of toast popups by @euvre in #19829
• fix(dataset): move enable/disable below run/cancel in bulk action menu by @dcc123456 in #19893
• fix(web): persist full embed dialog state in agent widget settings by @euvre in #19835
• Fix create build-in dataset parser_config should be pipeline config by @wangq8 in #19860
• fix: enhance PDF figure captions with vision by @buua436 in #19847
• Fix wording the Similarity threshold by @wangq8 in #19898
• Fix: mirror python when call llm to extract memory failed by @Lynn-Inf in #19863
• Expose KVROCKS for easy access by @wangq8 in #19904
• i18n: fill the remaining tr-TR gaps by @bakiburakogun in #19891
• fix(web): constrain flex children in agent version history dialog by @dcc123456 in #19903
• feat(model): fetch Xiaomi model list from remote API by @euvre in #19862
• fix(go-agent): surface CodeExec sandbox artifacts as message attachments by @mkaaad in #19852
• fix(agentic-rag): fail fast when the chat model cannot answer by @qinling0210 in #19850
• Go: fix ES connection port exhaustion under high concurrency by @JinHai-CN in #19901
• Fix: The picture and audio chunkers in the knowledge base display unnecessary configuration options. by @cike8899 in #19911
• fix(agent): host code-exec artifacts instead of leaking base64 into chat by @euvre in #19853
• Fix: expose rerank model error by @wangq8 in #19908
• refactor(go): drop dead KvrocksCheckpointExists probe by @xugangqiang in #19864
• fix(ingestion): route bare MinerU OCR model UUID to MinerU PDF dispatch by @euvre in #19639
• fix(web): surface memory-save failure as a toast in agent chat by @QAQadws in #19844
• Fix the build-in pipeline name by @wangq8 in #19921
• fix(api): enforce document image ownership authorization by @6ba3i in #19818
• Revert "fix(go-agent): surface CodeExec sandbox artifacts as message attachments" by @wangq8 in #19922
• Go: fix resource leak in ES connector by @JinHai-CN in #19912
• Fix: preserve MCP connection settings across import and edit by @georgeatparallel in #19518
• Go: fix other models http clients by @JinHai-CN in #19920
• fix(web): add missing audio and email images to dataset setting ImageMap by @dcc123456 in #19926
• fix(agent): surface invalid variables and deduplicate errors by @6ba3i in #19919
• Perf(deepdoc): free ONNX in/out tensors per Run instead of pinning them per session by @xugangqiang in #19843
• fix: prevent bulk overflow and preserve mindmap results by @buua436 in #19892
• fix(web): make audio attachments and references playable by @euvre in #19842
• fix(memory): stabilize filter option ordering by @6ba3i in #19925
• Removed CLAUDE.md since Claude Code supports AGENTS.md by @yuzhichang in #19932
• Make kvrocks host configurable by @wangq8 in #19916
• fix(service): use common.AssertURLSafe in mcp_test.go by @xugangqiang in #19931
• fix: Updated the pipeline tabs to show consistent labels. by @Yannnnnnny in #19933
• fix: release reused stemmer environments by @buua436 in #19935
• Fix: check SMTP config before send email by @Lynn-Inf in #19929
• Fix: Switching the dataset's embedding model causes the dropdown menu to disappear. by @cike8899 in #19927
• Fix re-parse complete document need pass delete:true for go by @wangq8 in #19936
• fix(web): remount chunk virtual list when result set changes by @dcc123456 in #19940
• refactor(go-ingestion): remove checkpoint/resume support by @xugangqiang in #19924
• fix(api): validate memory search numeric filters by @Harsh23Kashyap in #19883
• Fix dataset logs leak by @wangq8 in #19942
• fix(go): align retrieval and chat pipeline with Python by @qinling0210 in #19934
• Fix: hide MinerU.net by @Lynn-Inf in #19943
• fix(config): read boolean env flags so that off means off by @L4XB in #19689
• feat(agentic_rag): port the harness table/metadata paths to Go, with field-driven metadata_search by @Caobaining1 in #19941
• Fix: parse negative coordinates in @@ position tags by @xugangqiang in #19945
• chore(server): document missing CLI and MCP options by @6ba3i in #19900
• fix: stop SSE stream after terminal error by @midilandlu in #19870
• fix: handle empty final tool answers by @midilandlu in #19872
• fix: return only cited Go chat documents by @buua436 in #19899
• fix(deepdoc): route PDF parse logs through zap and drive colspan e2e in-process by @xugangqiang in #19947
• fix(agent): preserve empty switch input evaluation by @6ba3i in #19953
• fix(parser): keep media page-cache eviction alive after page-order break by @xugangqiang in #19938
• Fix: add a transient error, to retry by @wangq8 in #19956
• Fix/minio missing file error message by @QAQadws in #19918
• fix: honor Go chat citation settings and preserve final streamed citations by @QAQadws in #19797
• feat(models): add recent Gemini models by @Anurag-M1 in #19885
• Fix(codeexec): codeexec artifact attachments by @mkaaad in #19954
• fix(parser): scope TCADP bearer token to API server host only by @Taranum01 in #17672
• fix(retrieval): align Go parent-child chunk behavior by @jay77721 in #19828
• fix(parser): stop inlining PDF base64 media; crop on demand at VLM and chunker by @xugangqiang in #19964
• fix(ingestion): honor the dataset language in the Go parser by @jay77721 in #19966
• Go: update go.mod by @JinHai-CN in #19963
• Go CLI: fix benchmark by @JinHai-CN in #19946
• feat(parser): upgrade PDF header/footer removal with 3D cascaded fusion by @jay77721 in #19957
• refactor(parser): drop write-only setup fields from picture/audio parsers by @jay77721 in #19971
• fix(string_utils): keep spaces between non-Latin words by @chuenchen309 in #17015
• Port PR17015 to GO by @qinling0210 in #19969
• docs: add macOS notes to the launch-from-source guide by @JHWu1117 in #19970
• Fix: to add dataset logs by @wangq8 in #19973
• fix: align agentbot release flag handling by @buua436 in #19968
• fix(web): hide ingestion canvas test run button on Go backend by @jay77721 in #19977
• fix(pipeline): ignore disconnected nodes during terminal detection by @6ba3i in #19976
• fix(dataset): preserve parser config on pipeline switch by @QAQadws in #19978
• feat(ingestion): report sub-component parsing progress instead of a frozen 20% by @jay77721 in #19965
• feat(agentic_rag): put membership and document metadata into the answer's evidence by @Caobaining1 in #19981
• fix(parser): route UI "Plain Text" PDF method to the plain-text parser by @jay77721 in #19979
• Fix Agent cited not show during chat by @wangq8 in #19986
• fix(agent): normalize Switch operators and numeric strings by @6ba3i in #19987
• fix(tag): rebuild tagging on a Go tag source index by @qinling0210 in #19985
• fix(memory): pin memory filter facet options to a canonical order by @euvre in #19906
• chore: update stale CLAUDE.md references after the AGENTS.md rename by @Jinzhengxu in #19990
• Go CLI: show hardware by @JinHai-CN in #19991
• fix(embedding): cut inputs to the model window on every path by @yuzhichang in #19997
• Fix: Add a checklist to the agent canvas to display errors found on the canvas. by @cike8899 in #20000
• Fix save chilren delimiters by @wangq8 in #19998
• Fix: restrict metadata field key by @Lynn-Inf in #19999
• Go CLI: set/show cpu/memory/concurrency by @JinHai-CN in #20002
• Refactor: default log level to warn by @wangq8 in #20001
• fix(dataset): return accessible datasets for ids filter by @QAQadws in #19984
• Go: add deepdoc server framework by @JinHai-CN in #19988
• refactor(mcp): centralize configuration and transport wiring by @6ba3i in #20011
• feat(connector): add Zotero as a data source by @Souravrajvi0 in #19742
• fix(markdown): keep a pipe in a table cell from adding a column by @L4XB in #19749
• Go CLI: fix show user by @JinHai-CN in #20021
• Fix wordnet concurrency panic error by @yingfeng in #20023
• refactor(harness): extract retrieval wiring from server startup by @6ba3i in #20012
• Fix: delete document, does not delete the real file in minio by @wangq8 in #20022
• fix(parser-config): preserve child chunk settings on save by @6ba3i in #20018
• fix(ingestion, web): converge cancels fast and stop logging them as errors by @jay77721 in #19992
• Fix: delete datasets does not delete bucket and files in minio by @wangq8 in #20024
• fix(agent): stamp PDF watermark on every page of DocsGenerator exports by @euvre in #19915
• fix: synchronize document compilation and navigation projections by @buua436 in #20030
• Go: fix log level by @JinHai-CN in #20036
• fix(canvas): clarify unmatched Switch routing errors by @6ba3i in #20038
• fix: rename parserMethod labels by @dcc123456 in #20020
• fix(compilation): show update/log entry on Go backend for running and failed compiles by @dcc123456 in #20037
• fix: render Manual Chunker config in dataset parser tabs by @dcc123456 in #20043
• fix(agent): flag node groups detached from Begin in canvas checklist by @dcc123456 in #20046
• fix(agent): filter dragged next-step menu by source and skip empty menu by @dcc123456 in #20049
• fix(agent): do not create node when dragging from an input handle by @dcc123456 in #20052
• fix: prevent agent message template feedback by @buua436 in #20042
• refa: remove standalone tool support resolver by @buua436 in #20047
• refactor(go): unify Go logging on the zap logger by @jay77721 in #20026
• feat(deepdoc): make inference concurrency configurable (default 4) by @xugangqiang in #20044
• Fix: add hint url for local AI by @Lynn-Inf in #20039
• feat(deepdoc): share ONNX model weights across pooled native sessions by @xugangqiang in #20035
• Feature: delete user all stuffs when delete user by @wangq8 in #20010
• fix(api): align Go proxy route contracts by @6ba3i in #20032
• fix(document): route image and presentation uploads to their dedicated parsers by @jay77721 in #20061
• fix(ingestion): derive component param whitelist from component schemas by @jay77721 in #20060
• fix(parser): catch tight page numbers and site-promo headers/footers by @jay77721 in #20019
• refactor(parser): drop unreachable headerColEnd fallbacks from XLSX table items by @jay77721 in #20053
• refa: remove chat model target resolver by @buua436 in #20055
• fix(qa): read the value Excel computed, not the formula behind it by @L4XB in #19737
• fix(parser): replace slog.Debug with common.Debug+zap in toc_header_footer by @jay77721 in #20071
• perf(deepdoc): add single-core PDF throughput + peak-RSS benchmark by @xugangqiang in #20034
• fix(chunk): scope the chunk list and detail APIs to the route's datas… by @qinling0210 in #20059
• Revert "fix(api): align Go proxy route contracts" by @wangq8 in #20075
• fix(deepdoc): align OCR-rec batching with Python (sort by ratio, <=16, 96*1 pool) by @xugangqiang in #20041
• Fix: move migrate into init flag case list by @Lynn-Inf in #20048
• Fix Connecting other operators after the message operator is prohibited. by @cike8899 in #20076
• Fix unit test by @wangq8 in #20081
• Fix to add font to support generate English PDF by @wangq8 in #20078
• fix(api): add missing comma delimiter between question and extracted keywords by @ming1523 in #19661
• Port PR19661 to go (keyword delimiter) by @qinling0210 in #20069
• build(sh): pin ONNX Runtime archive sha256 instead of fetching it online by @xugangqiang in #20082
• fix(parser): drop eager base64 table image from pdf.TableItem by @xugangqiang in #20051
• fix(api): validate connector create payload by @Harsh23Kashyap in #19472
• chore(deps-dev): bump immutable from 4.3.7 to 4.3.9 in /web by @dependabot[bot] in #20086
• Update rag_analyzer.cpp (align with PR3503 in Infinity) by @qinling0210 in #20083
• test(rag): cover file magic-byte helpers in file_utils by @Bruce-Yii in #20072
• test(model_utils): add comprehensive tests for get_model_type_human and related functions by @Bruce-Yii in #20068
• chore(deps): bump anyio from 4.9.0 to 4.14.2 in /agent/sandbox by @dependabot[bot] in #20085
• chore(deps): bump soupsieve from 2.8.4 to 2.9 by @dependabot[bot] in #20087
• Fix no delete skill since go has no skill by @wangq8 in #20094
• fix(chat): prevent blank reference image preview when image url is not ready by @dcc123456 in #20097
• Fix: Add a loading state to the button that creates a memory. by @cike8899 in #20098
• chore(deps): bump go.opentelemetry.io/otel/sdk from 1.44.0 to 1.45.0 by @dependabot[bot] in #20088
• test(term_weight): add edge case tests for _alphabetic_oov_frequency by @Bruce-Yii in #20066
• chore(deps): bump anyio from 4.12.1 to 4.14.2 in /tools/es-to-oceanbase-migration by @dependabot[bot] in #20104
• fix: add standalone CJK font for Go PDF generation by @buua436 in #20095
• CI: fix kvrocks port is already allocated by @JinHai-CN in #20093
• Fix admin services display by @wangq8 in #20099
• Go CLI: fix --version by @JinHai-CN in #20112
• fix(markdown): read the first row of a header-less table as its header by @L4XB in #19750
• Fix: Restrict the canvas to a single compilation operator. by @cike8899 in #20120
• fix(agent): preserve content semantics with mixed Message branches by @Bruce-Yii in #19565
• fix(deepdoc): scope table/box overlap to the same page to avoid cross-page expansion by @xugangqiang in #20089
• chore(deps): bump anyio from 4.12.1 to 4.14.2 by @dependabot[bot] in #20105
• fix: scope document deletion compile events by @buua436 in #20108
• Fix: If the extractor is not configured with a model, a prompt is displayed in the checklist. by @cike8899 in #20113
• Fix: The configuration for the pipeline manual chunker cannot be displayed. by @cike8899 in #20106
• fix: wait for Elasticsearch chunk updates by @buua436 in #20116
• Fix delete dataset not delete bucket by @wangq8 in #20110
• fix: use configured known_hosts path for SSH sandbox by @buua436 in #20114
• Fix: Add the pipeline's detached operators to the checklist. by @cike8899 in #20127
• Fix: Hide the mind map update button for the Go version. by @cike8899 in #20130
• fix(chunk): support image updates on the chunk PATCH endpoint by @jay77721 in #20107
• Fix: rollback added instance when add model failed by @Lynn-Inf in #20129
• fix(agent): make begin form mode options reactive to language switch by @dcc123456 in #20128
• perf(deepdoc): bucket substring-dedup scan by page and column by @xugangqiang in #20122
• fix(deepdoc): keep highest-indexed table in MarkNoMergeTables page scoping by @xugangqiang in #20125
• fix(storage): correct Azure SPN object existence checks by @AnnaSuSu in #20003
• fix(markdown): delimit a link destination that holds a space or a parenthesis by @L4XB in #19747
• fix(chunker): eliminate O(n^2) coordinate extension and redundant merge copies by @xugangqiang in #20132
• fix(deepdoc): bucket table boxes by page in buildTableHTMLs by @xugangqiang in #20133
• i18n(az): add Azerbaijani (Azərbaycanca) translation by @jamalkamaladdin in #20131
• fix(service): drop citations and reference when the answer reports no… by @Caobaining1 in #20111
• Fix: not drop uncreated unique index by @Lynn-Inf in #20118
• fix(parser): accept longer closing fences during table extraction by @imexlovery in #20028
• fix(parser): split JSONL records only on newline characters by @imexlovery in #20029
• fix(deps): use numpy 2.x wheels on Python 3.13 (#19928) by @Souravrajvi0 in #20014
• i18n: sync tr-TR with recent en changes by @bakiburakogun in #20004
• Fix: clean ingestion task logs by @wangq8 in #20153
• fix: resolve go CLI latest release without GitHub API by @Yannnnnnny in #20159
• Go CI: fix port conflicts by @JinHai-CN in #20157
• docs: drop KIBANA_USER / KIBANA_PASSWORD, which nothing reads by @lihongyuan99 in #20150
• fix: update doc by @JinHai-CN in #20164
• fix(dataset): read document-declared tag sources in tag aggregation by @jay77721 in #20119
• fix: exclude disabled documents from Wiki products by @buua436 in #20154
• Fix rerank model does not take effect when search in searchbot by @wangq8 in #20158
• Fix eval error by @wangq8 in #20180
• Fix: when rerank error stop chat by @wangq8 in #20178
• Fix no regenerate/delete message in multiple model chats by @wangq8 in #20182
• Fix: dedup model name by @Lynn-Inf in #20167
• fix(parser): label PDF table items "table" only when they carry markup by @devYRPauli in #20092
• refactor(agent): pass checkpoint collaborators directly by @6ba3i in https://github.com/infiniflow/ragflow/pull/20168
• fix minimax model ids by @QAQadws in https://github.com/infiniflow/ragflow/pull/20184
• refactor(agent): remove unused canvas stream abstraction by @6ba3i in https://github.com/infiniflow/ragflow/pull/20169
• refactor(agent): remove obsolete component fixtures by @6ba3i in https://github.com/infiniflow/ragflow/pull/20170
• refactor(agent): remove unreachable canvas helpers by @6ba3i in https://github.com/infiniflow/ragflow/pull/20171
• refactor(agent): remove dead document io scaffolding by @6ba3i in https://github.com/infiniflow/ragflow/pull/20172
• refactor(agent): use concrete canvas state access by @6ba3i in https://github.com/infiniflow/ragflow/pull/20173
• fix(agent): reject malformed parallel checkpoints by @6ba3i in https://github.com/infiniflow/ragflow/pull/20174
• refactor(agent): remove dead switch map alias by @6ba3i in https://github.com/infiniflow/ragflow/pull/20175
• fix(agent): preserve standard JSON encodings by @6ba3i in https://github.com/infiniflow/ragflow/pull/20176
• fix(setting): suppress PaddleOCR form autofill and default URL to first option by @dcc123456 in https://github.com/infiniflow/ragflow/pull/20189
• feat(setting): auto-merge newly listed models into draft instance by @dcc123456 in https://github.com/infiniflow/ragflow/pull/20177
• feat(setting): default tool calling on for new provider instances by @dcc123456 in https://github.com/infiniflow/ragflow/pull/20195
• Fix: Switching the parsing mode when creating a dataset will cause the embedding model to change. by @cike8899 in https://github.com/infiniflow/ragflow/pull/20198
• fix(nav): navigation tree search, LLM cluster naming, cluster depth by @qinling0210 in https://github.com/infiniflow/ragflow/pull/20188
• perf(chunker): parallelize cropImageChunks across chunks by @xugangqiang in https://github.com/infiniflow/ragflow/pull/20183
• Fix rerank not working in agent retrieval by @wangq8 in https://github.com/infiniflow/ragflow/pull/20197
• Fix: The checklist allows loop operators to reference the outputs of internal operators. by @cike8899 in https://github.com/infiniflow/ragflow/pull/20205
• Fix: Document Generator – Header Text/Footer Text/Watermark Text configurations cannot be saved. by @cike8899 in https://github.com/infiniflow/ragflow/pull/20187
• fix(dataset): poll ingestion log while parsing and stop after terminal by @dcc123456 in https://github.com/infiniflow/ragflow/pull/20206
• Fix delete user to delete bucket user-downloads, table: file_commit, file_commit_item by @wangq8 in https://github.com/infiniflow/ragflow/pull/20199
• feat(chunker): stream cropped previews to storage instead of batching by @xugangqiang in https://github.com/infiniflow/ragflow/pull/20165
• fix(router): accept beta tokens on document image routes for shared chats by @jay77721 in https://github.com/infiniflow/ragflow/pull/20185
• Feat: Display wiki-related icons on the dataset's log page. by @cike8899 in https://github.com/infiniflow/ragflow/pull/20212
• Doc: add database change log by @Lynn-Inf in https://github.com/infiniflow/ragflow/pull/20211
• fix(dataset): confirm before re-parse Go documents with chunks by @dcc123456 in https://github.com/infiniflow/ragflow/pull/20210
• Fix: Switching from the dataset's "Files" page to the "Configuration" page results in the scrollbar not appearing at the top. by @cike8899 in https://github.com/infiniflow/ragflow/pull/20179
• fix: adjust sandbox SSH host configuration by @buua436 in https://github.com/infiniflow/ragflow/pull/20155
• fix(dataflow): make_encode_batch synchronous so thread_pool_exec can run it by @wuisabel-gif in https://github.com/infiniflow/ragflow/pull/16760
• fix(parser): read the separator a CSV was written with by @L4XB in https://github.com/infiniflow/ragflow/pull/19740
• fix(parser): fail an EPUB with no readable chapter instead of returning nothing by @L4XB in https://github.com/infiniflow/ragflow/pull/19814
• fix(agent): harden runtime boundary contracts by @6ba3i in https://github.com/infiniflow/ragflow/pull/20186
• fix: support wiki map writes on Infinity by @buua436 in https://github.com/infiniflow/ragflow/pull/20203
• Fix: Hide the 'local' option in the sandbox configuration. by @cike8899 in https://github.com/infiniflow/ragflow/pull/20269
• fix(tools): guard temporary table creation during mysql_migration dry run (#19570) by @Ritiky23 in https://github.com/infiniflow/ragflow/pull/19623
• feat: add Uzbek language support by @senovr in https://github.com/infiniflow/ragflow/pull/20264
• fix(parser): escape DOCX table cells in the naive, manual and qa chunkers by @L4XB in https://github.com/infiniflow/ragflow/pull/19746
• fix(chunker): skip crop+upload for non-retrievable empty-text media by @xugangqiang in https://github.com/infiniflow/ragflow/pull/20213
• Fix go workspaces APIs by @wangq8 in https://github.com/infiniflow/ragflow/pull/20272
• Go: remove unused files by @JinHai-CN in https://github.com/infiniflow/ragflow/pull/20271
• [codex] Fix OpenAI-compatible ASR provider by @HUAN2022A in https://github.com/infiniflow/ragflow/pull/19370
• fix(ingestion): read Python logs in Go backend by @jay77721 in https://github.com/infiniflow/ragflow/pull/20214
• fix(knowledge_compile): make compiled rows writable to Infinity by @qinling0210 in https://github.com/infiniflow/ragflow/pull/20208
• fix(ui): ignore outside clicks right after dialog opens by @dcc123456 in https://github.com/infiniflow/ragflow/pull/20270
• fix(pdf): rejoin line-break hyphens / strip U+FFFD in Go layout (temporary workaround) by @xugangqiang in https://github.com/infiniflow/ragflow/pull/20274
• Fix: It is not possible to select dataset metadata values ​​for the condition in the retrieval operator. by @cike8899 in https://github.com/infiniflow/ragflow/pull/20280
• disable python release schedule workflow by @myf-bee in https://github.com/infiniflow/ragflow/pull/20322
• PY: remove go related codes by @JinHai-CN in https://github.com/infiniflow/ragflow/pull/20366
• chore(deps): bump cryptography from 46.0.7 to 50.0.0 by @dependabot[bot] in https://github.com/infiniflow/ragflow/pull/20357
• Fix Python naive_merge dropping bare delimiters (#20276 parity) by @xugangqiang in https://github.com/infiniflow/ragflow/pull/20384
• fix(parser): match the .eml extension in any case by @devYRPauli in https://github.com/infiniflow/ragflow/pull/20090
• chore(deps): bump hpack from 4.1.0 to 4.2.0 by @dependabot[bot] in https://github.com/infiniflow/ragflow/pull/20356
• fix(nlp): flush buffer on bare delimiter in _build_cks (docx one-chunk regression) by @xugangqiang in https://github.com/infiniflow/ragflow/pull/20473
• fix(deepdoc): parse negative coordinates in @@ position tags on the Python side by @lihongyuan99 in https://github.com/infiniflow/ragflow/pull/19952
• fix(citation): open external web documents directly by @dcc123456 in https://github.com/infiniflow/ragflow/pull/20458
• feat(vastbase): add Vastbase G100 document engine support by @FengYikai710 in https://github.com/infiniflow/ragflow/pull/20064
• i18n(web): clarify agent citation switch label across locales by @dcc123456 in https://github.com/infiniflow/ragflow/pull/20608
• fix(llm): list inference-profile-only Bedrock models in API-key model discovery by @kimnamu in https://github.com/infiniflow/ragflow/pull/20163
• fix: report real token usage (chat + embedding) in chats_openai by @yangyachen in https://github.com/infiniflow/ragflow/pull/20031
• Fix: pin torch to a CUDA 12.x build to match onnxruntime-gpu by @iwasaki1112 in https://github.com/infiniflow/ragflow/pull/17030
• Fix: accept Ollama duration strings in OLLAMA_KEEP_ALIVE by @jclin0311 in https://github.com/infiniflow/ragflow/pull/20224
• fix(data_source): read Google OAuth env flags so that off means off by @lihongyuan99 in https://github.com/infiniflow/ragflow/pull/20009
• fix: honour chunk_token_num in the laws tree chunker by @shoemoney in https://github.com/infiniflow/ragflow/pull/18668
• fix(task): queue one PDF task for PaddleOCR, SoMark and TCADP by @devYRPauli in https://github.com/infiniflow/ragflow/pull/20136
• Fix: remove unnecessary files and rank conf in llm_factories.json by @Lynn-Inf in https://github.com/infiniflow/ragflow/pull/20681
• fix(webdav): tolerate Last-Modified clock skew in incremental poll by @leilei3167 in https://github.com/infiniflow/ragflow/pull/19967
• Bump Infinity to v0.7.4 by @qinling0210 in https://github.com/infiniflow/ragflow/pull/20664
• fix(api): validate top_k range for Dify retrieval endpoint by @Harsh23Kashyap in https://github.com/infiniflow/ragflow/pull/19436
• Bugfix : Restrict provider name by @wangq8 in https://github.com/infiniflow/ragflow/pull/21078
• Fix not restrict ids length when query datasets by @wangq8 in https://github.com/infiniflow/ragflow/pull/21089
• update env by @wangq8 in https://github.com/infiniflow/ragflow/pull/21118
• Update docker compose file by @wangq8 in https://github.com/infiniflow/ragflow/pull/21120
• Fix: The historical message window size for an agent operator must be an integer. by @cike8899 in https://github.com/infiniflow/ragflow/pull/21123
• Fix when parse with ASR model 1, no default model configure, report error; 2. model report error, should report error instead put error message as chunk content by @wangq8 in https://github.com/infiniflow/ragflow/pull/21125
• fix(parser): honor backtick-wrapped custom delimiter in txt parser by @jay77721 in https://github.com/infiniflow/ragflow/pull/21124
• fix(flow): sanitize numpy types in pipeline output before JSON serialization by @jay77721 in https://github.com/infiniflow/ragflow/pull/21121
• Fix: Remove headers and footers, and remove the original table of contents. Keep only PDF and DOCX files; delete all other file types. by @cike8899 in https://github.com/infiniflow/ragflow/pull/21133
• fix(pipeline): load persisted parsing components in Python by @6ba3i in https://github.com/infiniflow/ragflow/pull/21137
• fix(agent): allow incomplete drafts and clarify validation errors by @6ba3i in https://github.com/infiniflow/ragflow/pull/21138
• fix(parser): recognize the plain text PDF method by @6ba3i in https://github.com/infiniflow/ragflow/pull/21139
• fix(agent): hide Run button on ingestion pipeline page by @6ba3i in https://github.com/infiniflow/ragflow/pull/21148
• fix(web): only close rename modal after a successful save by @dcc123456 in https://github.com/infiniflow/ragflow/pull/21151
• Fix: Align the <th> elements within the parsed chunk to the left. by @cike8899 in https://github.com/infiniflow/ragflow/pull/21142
• fix(storage): fail fast instead of storing plaintext when crypto config is broken by @errmakov in https://github.com/infiniflow/ragflow/pull/20382
• fix(mws): handle tool-call history and prevent unbound tool execution by @nikminer in https://github.com/infiniflow/ragflow/pull/20307
• feat(llm): add Radeon Cloud provider support by @6ba3i in https://github.com/infiniflow/ragflow/pull/21154
• fix: not allow case-insensitive duplicate provider name by @Lynn-Inf in https://github.com/infiniflow/ragflow/pull/21165
• fix(chunking): apply configured overlap to DOCX text chunks by @jay77721 in https://github.com/infiniflow/ragflow/pull/21152
• Fix rereank model wrong key redirect to login page by @wangq8 in https://github.com/infiniflow/ragflow/pull/21168
• fix(chunking): preserve title hierarchy when splitting oversized chunks by @jay77721 in https://github.com/infiniflow/ragflow/pull/21167
• fix(wiki): recover transient page-writer failures after retry exhaustion by @ccclucky in https://github.com/infiniflow/ragflow/pull/20930
• fix(migration): backfill document.meta_fields into per-tenant metadata stores by @qinling0210 in https://github.com/infiniflow/ragflow/pull/21157
• Fix LLM setting for 3 search config (#20955) by @wangq8 in https://github.com/infiniflow/ragflow/pull/21197
• fix: remove KNN fallback from structure search by @buua436 in https://github.com/infiniflow/ragflow/pull/21195
• fix(chunk): write chunk_order_int on every ingestion path by @qinling0210 in https://github.com/infiniflow/ragflow/pull/21164
• Fix: knowledge base files cannot preview by @wangq8 in https://github.com/infiniflow/ragflow/pull/21200
• fix(qa): keep a comma CSV comma separated when one field holds a TAB by @devYRPauli in https://github.com/infiniflow/ragflow/pull/20925
• test: remove import-time module stubs that leak into later tests by @devYRPauli in https://github.com/infiniflow/ragflow/pull/20926
• test(rag): cover non-adjacent repeated cells in the DOCX table colspan loop by @devYRPauli in https://github.com/infiniflow/ragflow/pull/20927
• fix(parser): preserve escaped text in HTML tables by @imexlovery in https://github.com/infiniflow/ragflow/pull/20027
• fix: cap chunk list page size when filtering by chunk ids by @cike8899 in https://github.com/infiniflow/ragflow/pull/21207
• test(parser): cover JSONL records of 0 and false by @devYRPauli in https://github.com/infiniflow/ragflow/pull/20928
• Fix: On the multi-model comparison page, the "Apply," "Delete," and "Add" buttons should be disabled whenever any of the models is generating output. by @cike8899 in https://github.com/infiniflow/ragflow/pull/21224
• fix(web): bind memory message switch state to the message by @dcc123456 in https://github.com/infiniflow/ragflow/pull/21223
• Fix: Agent - Start Component - When an input variable is set to optional: if a value is entered and then deleted, clicking "Next" triggers a "required field" error. by @cike8899 in https://github.com/infiniflow/ragflow/pull/21225
• Fix agent generated files cannot download by @wangq8 in https://github.com/infiniflow/ragflow/pull/21231
• fix: record all dataset compilation logs by @buua436 in https://github.com/infiniflow/ragflow/pull/21227
• fix(rag): allow agentic answers with retrieved evidence on 0.27.x by @ccclucky in https://github.com/infiniflow/ragflow/pull/20647
• fix(deepdoc): surface empty page-batch results from VisionParser through the task callback by @Harsh23Kashyap in https://github.com/infiniflow/ragflow/pull/18650
• fix(deepdoc): normalize coordinates before crop to prevent crash by @ResidualBlood in https://github.com/infiniflow/ragflow/pull/20282
• fix(agent): allow shared datasets across template blocks by @cike8899 in https://github.com/infiniflow/ragflow/pull/21235
• fix(retrieval): drop the filter-only fallback by @qinling0210 in https://github.com/infiniflow/ragflow/pull/21242
• Fix when VLM model failure, task should fail by @wangq8 in https://github.com/infiniflow/ragflow/pull/21249
• Fix: sort memory messages by combined key by @Lynn-Inf in https://github.com/infiniflow/ragflow/pull/21246
• fix(api): do not reuse task chunks the doc store no longer has by @ebarkhordar in https://github.com/infiniflow/ragflow/pull/20191
• fix(web): accept null table_column_* fields in dataset settings form by @surlymochan in https://github.com/infiniflow/ragflow/pull/21110
• Fix deepseek VLM model does not work by @wangq8 in https://github.com/infiniflow/ragflow/pull/21253
• fix(llm): route Bedrock OpenAI GPT-5.x/GPT-6 through Converse and drop temperature/top_p/stop by @kimnamu in https://github.com/infiniflow/ragflow/pull/20161
• fix(api): return 400 for non-numeric searchbots retrieval_test parameters by @Harsh23Kashyap in https://github.com/infiniflow/ragflow/pull/19471
• fix: accept legacy positional task executor index by @helloxjade in https://github.com/infiniflow/ragflow/pull/16147
• fix(db): postgres in-place upgrade for model_type and tenant_*_id by @Souravrajvi0 in https://github.com/infiniflow/ragflow/pull/18781
• fix(rag/app): separate parse progress counts by @MarMarLabs in https://github.com/infiniflow/ragflow/pull/19586
• fix(agent): log validation failures and trim compat docstring by @kabishou11 in https://github.com/infiniflow/ragflow/pull/20201
• Fix: show all agent messages by @wangq8 in https://github.com/infiniflow/ragflow/pull/21264
• Fix: disable citation should not show references by @wangq8 in https://github.com/infiniflow/ragflow/pull/21268
• fix(api): fall back to create_time for an unknown orderby on the file list by @MarMarLabs in https://github.com/infiniflow/ragflow/pull/19958
• fix(api): pass the message by keyword in file-commit error replies by @MarMarLabs in https://github.com/infiniflow/ragflow/pull/19685
• fix(retrieval): scope mother chunks by document by @noooooooookro in https://github.com/infiniflow/ragflow/pull/19375
• fix(infinity): normalize *_kwd fields that Infinity returns as lists by @qinling0210 in https://github.com/infiniflow/ragflow/pull/21265
• Fix: check duplicate name when create and update instance by @Lynn-Inf in https://github.com/infiniflow/ragflow/pull/21272
• fix(opensearch): register 3072 and 3584 vector dimensions by @yosimasu in https://github.com/infiniflow/ragflow/pull/21280
• fix(chunking): split multiline records before title detection by @jay77721 in https://github.com/infiniflow/ragflow/pull/21270
• fix(deepdoc): normalize generated positions and rotated OCR page indices by @ResidualBlood in https://github.com/infiniflow/ragflow/pull/21254
• fix(docker): proxy /mcp to the MCP server in nginx by @yosimasu in https://github.com/infiniflow/ragflow/pull/21276
• fix(api): validate chunk available/available_int before int() (#19479) by @Taranum01 in https://github.com/infiniflow/ragflow/pull/19784
• fix(wiki): prevent nested progress from completing dataset tasks by @ccclucky in https://github.com/infiniflow/ragflow/pull/21296
• fix(wiki): generate wiki pages on Infinity instead of silent empty results by @qinling0210 in https://github.com/infiniflow/ragflow/pull/21290
• fix(agent): require explicit edges for DataOperations query refs by @kabishou11 in https://github.com/infiniflow/ragflow/pull/19698
• fix(structure-graph): page dataset merge past ES max_result_window by @errmakov in https://github.com/infiniflow/ragflow/pull/20400
• fix(qa): localize table pages for partial PDF ranges by @ResidualBlood in https://github.com/infiniflow/ragflow/pull/21299
• fix(agent): allow shared dataset metadata lookups by @6ba3i in https://github.com/infiniflow/ragflow/pull/21318
• Fix email parser by @wangq8 in https://github.com/infiniflow/ragflow/pull/21321
• fix(web): unmount form fields when shouldRender is false by @dcc123456 in https://github.com/infiniflow/ragflow/pull/21322
• fix(web): refetch agent log on reset (#17050) by @Taranum01 in https://github.com/infiniflow/ragflow/pull/20266
• fix(llm): unify api_base routing and fix Bedrock auth_mode compatibility by @211250105 in https://github.com/infiniflow/ragflow/pull/15826
• feat(mineru): support official mineru.net API by @huang-aoqin in https://github.com/infiniflow/ragflow/pull/19980
• fix(agent): route Categorize ties to the most specific category name by @Harsh23Kashyap in https://github.com/infiniflow/ragflow/pull/19411
• fix(infinity): split inserts above the 8192-row statement limit by @qinling0210 in https://github.com/infiniflow/ragflow/pull/21338
• fix(chunking): place PDF figures and tables in reading order by @qinling0210 in https://github.com/infiniflow/ragflow/pull/21337
• fix(agent): restore chat output after branch selection by @6ba3i in https://github.com/infiniflow/ragflow/pull/21378
• add release-python workflow by @myf-bee in https://github.com/infiniflow/ragflow/pull/21382
• Docs: Update version references to v0.27.3 in READMEs and docs by @asiroliu in https://github.com/infiniflow/ragflow/pull/21354

  New Contributors

  • @kabishou11 made their first contribution in #19206
  • @huyua9 made their first contribution in #18951
  • @aroh3006 made their first contribution in #19517
  • @wadefelix made their first contribution in #15998
  • @coderdailyone made their first contribution in #19543
  • @anonrouterai made their first contribution in #19561
  • @brucehu made their first contribution in #18598
  • @Ritiky23 made their first contribution in #19559
  • @spicyfishPro made their first contribution in #19483
  • @DennyHo0917 made their first contribution in #19566
  • @gluedtea made their first contribution in #19541
  • @lucasoffchain made their first contribution in #19419
  • @seven7763 made their first contribution in #19560
  • @ori-frenkel made their first contribution in #19407
  • @wang393 made their first contribution in #19521
  • @HUAN2022A made their first contribution in #17669
  • @ning-jl made their first contribution in #19498
  • @aiapienthusiast made their first contribution in #19748
  • @MQ-xz made their first contribution in #19817
  • @georgeatparallel made their first contribution in #19518
  • @Anurag-M1 made their first contribution in #19885
  • @JHWu1117 made their first contribution in #19970
  • @jamalkamaladdin made their first contribution in #20131
  • @imexlovery made their first contribution in #20028
  • @lihongyuan99 made their first contribution in #20150
  • @senovr made their first contribution in https://github.com/infiniflow/ragflow/pull/20264
  • @yangyachen made their first contribution in https://github.com/infiniflow/ragflow/pull/20031
  • @jclin0311 made their first contribution in https://github.com/infiniflow/ragflow/pull/20224
  • @shoemoney made their first contribution in https://github.com/infiniflow/ragflow/pull/18668
  • @ccclucky made their first contribution in https://github.com/infiniflow/ragflow/pull/20930
  • @ResidualBlood made their first contribution in https://github.com/infiniflow/ragflow/pull/20282
  • @surlymochan made their first contribution in https://github.com/infiniflow/ragflow/pull/21110
  • @noooooooookro made their first contribution in https://github.com/infiniflow/ragflow/pull/19375
  • @yosimasu made their first contribution in https://github.com/infiniflow/ragflow/pull/21280
  • @211250105 made their first contribution in https://github.com/infiniflow/ragflow/pull/15826

  Full Changelog: v0.27.2...v0.27.3