v0.21.0

Released on October 15, 2025.

New features

• Orchestratable ingestion pipeline: Supports customized data ingestion and cleansing workflows, enabling users to flexibly design their data flows or directly apply the official data flow templates on the canvas.
• GraphRAG & RAPTOR write process optimized: Replaces the automatic incremental build process with manual batch building, significantly reducing construction overhead.
• Long-context RAG: Automatically generates document-level table of contents (TOC) structures to mitigate context loss caused by inaccurate or excessive chunking, substantially improving retrieval quality. This feature is now available via a TOC extraction template.
• Admin CLI: Introduces a new command-line tool for system administration, allowing users to manage and monitor RAGFlow's service status via command line.

Improvements

• Redesigns RAGFlow's Login and Registration pages.
• Upgrades RAGFlow's document engine Infinity to v0.6.0.

Added models

• Tongyi Qwen 3 series
• Claude Sonnet 4.5
• Meituan LongCat-Flash-Thinking

New agent templates

• Company Research Report Deep Dive Agent: Designed for financial institutions to help analysts quickly organize information, generate research reports, and make investment decisions.
• Orchestratable Ingestion Pipeline Template: Allows users to apply this template on the canvas to rapidly establish standardized data ingestion and cleansing processes.

What's Changed

• Feat: dataflow supports Spreadsheet and Word processor document by @yongtenglei in #9996
• Feat: dataflow supports markdown by @Lynn-Inf in #10003
• Refactor:Improve BytesIO usage for image2base64 by @Woody-Hu in #9997
• Fix: Fixed the issue where newly added tool operators would disappear after editing the form #10013 by @cike8899 in #10016
• Feat: Add type card to create agent dialog #9869 by @cike8899 in #10025
• Feat(config): Update service config template new defaults by @asiroliu in #10029
• Feat: Import dsl from agent list page #9869 by @cike8899 in #10033
• feat: Added UI functions related to data-flow knowledge base #3221 by @dcc123456 in #10038
• Refactor:Improve BytesIO usage for GeminiCV by @Woody-Hu in #10042
• Feat: Translate the parser operator #9869 by @cike8899 in #10037
• Fix: In ragflow/rag/app /naive.py, if there are multiple images in one line, the other images will be lost by @Woody-Hu in #9968
• Fix: token num exceed by @Lynn-Inf in #10046
• Fix: Hide dataflow related functions #9869 by @dcc123456 in #10045
• The chat dialog box cannot be fully displayed on a small screen #10034 by @cike8899 in #10049
• Feat: add TokenPony model provider by @zhileihuang in #9932
• Docs: How to connect to an MCP server as a client by @writinwaters in #10043
• Feat: Agent component support inserting variables(#10048) by @wilmerxu in #10055
• Feat: dataflow supports text by @Lynn-Inf in #10058
• Feat: Translate the fields of the embedded dialog box on the agent page #3221 by @cike8899 in #10072
• Fix: terminate onnx inference session manually by @Lynn-Inf in #10076
• feat: add multiple docker tags (latest, latest_full, latest_slim) to … by @adrianweidig in #10040
• Fix: The same model appears twice in the drop-down box. #10102 by @cike8899 in #10103
• Fix: judge not empty before delete by @Lynn-Inf in #10099
• Feat: add support for the Ascend layout recognizer by @yongtenglei in #10105
• Docs: Added an Execute SQL tool reference by @writinwaters in #10108
• Fix(agent template): wrap template variables in curly braces by @asiroliu in #10109
• Feat: add support for the Ascend table structure recognizer by @yongtenglei in #10110
• Feat/parse img by @Lynn-Inf in #10112
• Fix: set default chunk_token_num in html_parser by @buua436 in #10118
• Fix: usage of postgresql -> postgres for db_type by @cucusenok in #10120
• Docs: Removed /v1 from Ollama base URLs by @writinwaters in #10067
• Fixes session_id passing in agent_openai completion. by @bluthen in #10124
• Feat: parse audio by @Lynn-Inf in #10133
• Build(deps): Bump axios from 1.9.0 to 1.12.0 in /sandbox/sandbox_base_image/nodejs by @dependabot[bot] in #10091
• Support image recognition for url links in Markdown file, fix log error in code_exec by @Magicbook1108 in #10139
• Correct the text of vectorSimilarityWeight in zh.ts by @aopstudio in #10128
• Docs: Updated database types supported by the Execute SQL tool by @writinwaters in #10113
• Feat: add CometAPI to LLMFactory and update related mappings by @TensorNull in #10119
• Feat: add support for KB document basic info by @yongtenglei in #10134
• Fix: embedding model failure in CometAPI by @buua436 in #10137
• Fix: handle zero by @Lynn-Inf in #10149
• Fix: Merge different types of models from the same manufacturer #10146 by @cike8899 in #10157
• Feat: migrate OpenAI-compatible chats to LiteLLM by @yongtenglei in #10148
• Fix: knowledge base's embedded model form layout and dependency imports in the main branch. #9869 by @dcc123456 in #10160
• Add Firecrawl integration for RAGFlow by @aybanda in #10152
• Feat: Add file convert to document API just like file2document_app.py by @huang-aoqin in #10158
• Support server health check by http://localhost:/v1/system/healthz by @Magicbook1108 in #10150
• Docs: Input and output variables defined in the Input and Output sections must also be implemented in your code. by @writinwaters in #10162
• fix: correct rerank_model condition logic by @buua436 in #10174
• Docs: How to accelerate question answering by @writinwaters in #10179
• Feat: add support for Anthropic third-party API by @buua436 in #10173
• perf: add get_all_kb_doc_count func to simplify kb.doc_num updating by @whhe in #10169
• Fix: skip tag query if tag kbs are invalid by @whhe in #10168
• Fix: enable scrolling at chat setting by @atsushi729 in #10184
• Feat: add admin CLI and admin service by @JinHai-CN in #10186
• Chat Widget by @melmathari in #10187
• Add tree_merge for law parsers, significantly outperforming hierarchical_merge by @Magicbook1108 in #10202
• Refactor: use the same implement for total token count from res by @Woody-Hu in #10197
• fix python_api example by @zhangwfjh in #10196
• The retrieval component needs to support returning JSON data(#10170) by @wilmerxu in #10171
• Fix: resolve hash collisions by switching to UUID &correct logic for always-true statements & Update GPT api integration & Support qianwen-deepresearch by @Magicbook1108 in #10208
• Fix: string merge bug about agent TextProcessing. (\m) by @zhimeng123 in #10211
• Fix bug: broken import from rag.prompts.prompts by @JinHai-CN in #10217
• Fix: azure OpenAI retry by @SgtMarmite in #10213
• revert gpt5 integration by @Magicbook1108 in #10228
• Flow text processing bug by @zhimeng123 in #10246
• fix: prevent list index out of range in chat streaming by @buua436 in #10238
• UI: Add Russian language by @rjohny55 in #10249
• Fix:max_tokens must be at least 1, got -950, BadRequestError by @Woody-Hu in #10252
• UI: Update Russian language ru.ts by @mcoder6425 in #10251
• Fix: Russian language config.ts by @rjohny55 in #10250
• add Qwen models by @Magicbook1108 in #10263
• add model: qwen3-max and qewn3-vl series by @TeslaZY in #10256
• fix: Wrong Qwen models's ID by @TeslaZY in #10272
• Docs: Knowledge base renamed to dataset. by @writinwaters in #10269
• Refactor: Improve the logic clean conf for ZhipuChat by @Woody-Hu in #10274
• fix: replace traceback.print_exc() with logging.exception(e) in conve… by @buua436 in #10275
• Refactor: fix typos by @JinHai-CN in #10200
• Refactor: move encrypt/decrypt to one file by @JinHai-CN in #10203
• Fix: invalid COMPONENT_EXEC_TIMEOUT by @yongtenglei in #10278
• Feat/admin service by @Lynn-Inf in #10233
• Feat(nlp): add "怎么办" pattern to question word removal by @asiroliu in #10284
• Fixed the issue where database connections were interrupted under high concurrency by @lemsn in #10126
• Refactor: move some functions out of api/utils/init.py by @JinHai-CN in #10216
• Refactor: migrate the function to specific file by @JinHai-CN in #10201
• FIX: Rename the CometEmbed and CometSeq2txt classes to CometAPIEmbed and CometAPISeq2txt, and correct supported_models.mdx. by @TensorNull in #10298
• feat: add IBM DB2 support by @buua436 in #10306
• Fix:After setting user's is_active to 0, the user can still log in to RAGFlow. by @Woody-Hu in #10325
• Fix: The enterprise version of the knowledge graph cannot be displayed. #10333 by @cike8899 in #10334
• fix: remove ibm-db dependency and refactor import order by @buua436 in #10330
• Fix: floating widget match style with original one by @adagora in #10317
• Fix: update Japanese translations in ja.ts by @atsushi729 in #10338
• Fix: The dataset uses the new id to obtain the knowledge graph #10333 by @cike8899 in #10339
• fix: Handling Null Values in SQL Execution Results by @Womsxd in #10332
• Fix: unexpected operation of file management by @yongtenglei in #10343
• Feat/admin drop user by @Lynn-Inf in #10342
• Fix: invalid user can login from OSS by @Woody-Hu in #10348
• Feat: Keep connection status during generating agent by drag and drop … by @FatMii in #10141
• Feat: Add baseUrl to the Tongyi Qianwen model configuration modal #10340 by @cike8899 in #10357
• Feat: add support for international Dashscope service by @yongtenglei in #10356
• Feat: add support for LongCat-Flash-Thinking and Claude Sonnet 4.5 by @yongtenglei in #10374
• Refactor: move exceptions to common folder by @JinHai-CN in #10383
• Fix: reset tools in/out-put. by @KevinHuSh in #10378
• Fix: unexpected operation of document management by @yongtenglei in #10366
• fix: reset chat state when creating new dialog by @buua436 in #10380
• Feat/service manage by @Lynn-Inf in #10381
• Doc: Fixed a typo by @writinwaters in #10391
• Fix highlight with infinity by @yuzhichang in #10345
• Check workflow duplication by @yuzhichang in #10399
• Fix MY_GITHUB_TOKEN by @yuzhichang in #10404
• Refactor: improve cv model logics by @Woody-Hu in #10414
• Feature/agent UI style optimization by @FatMii in #10385
• Feat: add DeerAPI support by @DeerAPI-dev in #10303
• Refactor: fix admin exception by @JinHai-CN in #10400
• Feat: Use data pipeline to visualize the parsing configuration of the knowledge base by @KevinHuSh in #10423
• Feat: Adds a new feature that enables the LLM to extract a structured table of contents (TOC) directly from plain text. by @Magicbook1108 in #10428
• Doc: guide for admin service by @Lynn-Inf in #10431
• Feat: Modify the translation file of the workflow #9869 by @cike8899 in #10430
• Feat: HTTP componant supports variables by @yongtenglei in #10432
• Fix: Bug fixes and removed previous settings page code #9869 by @dcc123456 in #10435
• fix:update searxng_url logic by @buua436 in #10440
• Feat: Replace the collapse icon #9869 by @cike8899 in #10442
• Fix: correctly update parser method & correct vllm pdf parser by @Magicbook1108 in #10441
• Feat: Importing data flow files from the list page #9869 by @cike8899 in #10446
• Fix typos: retrievaler -> retriever by @JinHai-CN in #10372
• Feat: Pipeline Docx file supports Markdown output by @yongtenglei in #10439
• Fix OCR GPU provider mem limit handling by @loks666 in #10407
• Refactor: improve how NvidiaCV calculate res total token counts by @Woody-Hu in #10455
• Feat: add total in List dataset API by @Magicbook1108 in #10448
• Feat: Add prompts for toc relevance check according to #10436 by @Magicbook1108 in #10457
• bump infinity by @yuzhichang in #10422
• Fix: Fixed the issue where the operator added by clicking the plus sign in the data flow would overlap with the original section #9886 by @cike8899 in #10458
• Fix: Fixed an issue where parser configurations could be added infinitely #9869 by @cike8899 in #10464
• feat: support qwq reasoning models with non-stream output by @buua436 in #10468
• Feat: TOC retrieval by @KevinHuSh in #10456
• Fix: typo by @Lynn-Inf in #10469
• Fix: Fixed some discovered bugs #9869 by @dcc123456 in #10466
• Feat: Added toc enhance field to chat and retrieval operator configuration #10436 by @cike8899 in #10470
• typo: fix _entity_index_dilimiter_key to entity_index_delimiter_key by @ffffffff0 in #10465
• Fix: Fixed the issue where swagger apidocs could not be opened properly(#9522) by @yngvarhuang in #10461
• Fix #10408 by @glu000 in #10471
• Feat: Adjust the style of the delete button on the agent side #9869 by @cike8899 in #10473
• Fix: Agent.reset() argument wrong #10463 & Unable to converse with agent through Python API. #10415 by @Magicbook1108 in #10472
• fix: return type annotation for get_urls() in download_deps by @gnowland in #10478
• Fix: Disable reasoning on Gemini 2.5 Flash by default by @glu000 in #10477
• Fix: Fixed the issue where the connection lines of placeholder nodes in the agent canvas could not be displayed #9869 by @cike8899 in #10485
• feat(login): Refactor the login page and add dynamic background and highlight effects #9869 by @dcc123456 in #10482
• Feat: include author, journal name, volume, issue, page, and DOI in PubMed search results by @buua436 in #10481
• Feat: debugging toc part. by @KevinHuSh in #10486
• Feat: Modify the default style of the agent node anchor #9869 by @cike8899 in #10489
• Fix: Added table of contents extraction functionality and optimized form item layout #9869 by @dcc123456 in #10492
• Feat: add support for multi-column PDF parsing by @yongtenglei in #10475
• Fix: incorrect agent template #10393 by @Magicbook1108 in #10491
• Fix: remove lang for autio. by @KevinHuSh in #10496
• Feat: ​​OpenSearch's support for newly embedding models​​ by @pyyuhao in #10494
• Fix: Mcp reset error, #10497 by @Magicbook1108 in #10498
• Feat: Modify the background color of the canvas #9869 by @cike8899 in #10507
• fix: decode before format to json by @Lynn-Inf in #10506
• Fix: maintain backward compatibility for KB tasks by @yongtenglei in #10508
• Feat: add trino support by @buua436 in #10512
• Fix: Update lm studio models support, refer to #8116 by @Magicbook1108 in #10509
• Fix: The Context Generator(Transformer) node can only be followed by a Tokenizer(Indexer) and a Context Generator(Transformer). #9869 by @cike8899 in #10515
• Fix: Optimized the login page and fixed some known issues. #9869 by @dcc123456 in #10514
• Feat: Merge title splitter and token splitter into chunker category #9869 by @cike8899 in #10517
• Docs: Updated descriptions on metadata filtering by @writinwaters in #10518
• Fix: XSS vulnerability in Ragflow's chat view by @cike8899 in #10519
• Feat: add admin server to docker by @Lynn-Inf in #10522
• Fix: wrong Knowledgebase tasks_finish_at by @yongtenglei in #10521
• Feat: Add parse_document with feed back by @Magicbook1108 in #10523
• Fix: empty references in agent conversation by @buua436 in #10528
• Feat: add Zhipu GLM-ASR model by @yongtenglei in #10529
• Add kibana tool in the docker compose file(#10525) by @yngvarhuang in #10526
• Feat: Display the configuration of data flow operators on the node #9869 by @cike8899 in #10533
• Fix: Update the parsing editor to support dynamic field names and optimize UI styles #9869 by @dcc123456 in #10535
• Add 'status' field when list services by @JinHai-CN in #10538
• Docs: Updated parse_documents by @writinwaters in #10536
• Fix: unexpected Auth return code by @yongtenglei in #10539
• Docs: The full edition has only two built-in embedding models by @writinwaters in #10540
• Feat: let toc run asynchronizly... by @KevinHuSh in #10513
• The 'cmd' module is introduced to make the CLI easy to use. by @JinHai-CN in #10542
• Feat: Merge splitter and hierarchicalMerger into one node #9869 by @cike8899 in #10543
• Fix: update parser id for change_parser. by @KevinHuSh in #10545
• Feat: build ragflow-cli by @Lynn-Inf in #10544
• Feat: Adjust the style of note nodes #9869 by @cike8899 in #10547
• Fix: Switch the default theme from light mode to dark mode and improve some styles #9869 by @dcc123456 in #10552
• Fix: When I click to interrupt the chat, the page reports an error #10553 by @cike8899 in #10554
• Fix: canvas list with team. by @KevinHuSh in #10549
• bump infinity to v0.6.0 by @yuzhichang in #10558
• Feat: auto release by @Lynn-Inf in #10557
• Fix: Adding Ingestion Pipeline Classification to Agents Template #9869 by @dcc123456 in #10556
• Feat: Contribute ingestion pipeline templates by @ZhenhangTung in #10551
• Feat: Create Stock_research_report.json by @Llala0928 in #10555
• Doc: Added v0.21.0 release notes by @writinwaters in #10559
• Fix release notes by @JinHai-CN in #10560
• Update README by @KevinHuSh in #10563
• Fix: agent templates... by @KevinHuSh in #10564
• Fix: Click the reset button on the agent page shared externally, and the greeting in conversation mode should not be deleted. #10567 by @cike8899 in #10571
• Docs: Update version references to v0.21.0 in READMEs and docs by @asiroliu in #10565
• Feat: Creating a data flow from a template page #9869 by @cike8899 in #10573
• Fix: Optimize metadata filters, add Ingestion pipeline options to agent templates page #9869 by @dcc123456 in #10572
• Fix admin: can't read config and empty line error by @JinHai-CN in #10574

New Contributors

• @zhileihuang made their first contribution in #9932
• @wilmerxu made their first contribution in #10055
• @adrianweidig made their first contribution in #10040
• @buua436 made their first contribution in #10118
• @cucusenok made their first contribution in #10120
• @bluthen made their first contribution in #10124
• @Magicbook1108 made their first contribution in #10139
• @TensorNull made their first contribution in #10119
• @aybanda made their first contribution in #10152
• @atsushi729 made their first contribution in #10184
• @melmathari made their first contribution in #10187
• @zhangwfjh made their first contribution in #10196
• @zhimeng123 made their first contribution in #10211
• @SgtMarmite made their first contribution in #10213
• @mcoder6425 made their first contribution in #10251
• @lemsn made their first contribution in #10126
• @adagora made their first contribution in #10317
• @DeerAPI-dev made their first contribution in #10303
• @loks666 made their first contribution in #10407
• @ffffffff0 made their first contribution in #10465
• @yngvarhuang made their first contribution in #10461
• @glu000 made their first contribution in #10471
• @ZhenhangTung made their first contribution in #10551
• @Llala0928 made their first contribution in #10555

Full Changelog: v0.20.5...v0.21.0