infiniflow/ragflow v0.21.0 on GitHub

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.
Video file parsing: Expands the system's multimodal data processing capabilities by supporting video file parsing.
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