v0.8.2

eosphoros-ai/DB-GPTv0.8.2Aug 26, 2026by github-actions[bot]

AI Summary

This release expands data analysis capabilities with multi-file upload and MaxCompute support, while enhancing agent workflows through parallel delegation and knowledge-base search. It also prioritizes security by addressing critical vulnerabilities like RCE and path traversal.

Key Highlights

  • Multi-file upload and analysis support
  • Agentic Knowledge-Base Search with indexing and RAG
  • Parallel sub-agent delegation execution
  • MaxCompute (ODPS) datasource integration
  • Critical security fixes for RCE and path traversal

Breaking Changes

  • API method change: `create_datasource` and `create_flow` endpoints now require POST requests instead of GET.
  • Agent output format change: Citations are now returned separately from the final answer.
  • Skill file naming restriction: Only files named `SKILL.md` are loaded as Markdown skills.

New Features

  • Multi-file upload and analysis
  • OrcaRouter proxy provider
  • MaxCompute (ODPS) datasource support
  • Parallel sub-agent delegation execution
  • Agentic Knowledge-Base Search (Indexing + Agentic RAG)
  • Built-in `ask_question` tool

Full Release Notes

## ✨ Enhancements

- support multi-file upload and analysis  (#3206)
- add OrcaRouter proxy provider (#3186)
- add MaxCompute (ODPS) datasource support (#3105) (#3178)
- Parallel sub-agent delegation execution (#3161)
- Agentic Knowledge-Base Search (Indexing + Agentic RAG) (#3160)
- add ask_question tool as builtin-tools (#3107)

## 🐞 Bug fixes

- avoid KeyError when chart/sql run omits db\_name (#3199)
- include flow uid in update\_flow PUT path (#3193) (#3196)
- use LONGTEXT for large agent messages on MySQL (#3189)
- move EXAMPLE\_1 database creation to end of schema file (#3183)
- separate citations from final answers (#3182)
- validate user\_id to prevent path traversal in python file upload (#3104) (#3184)
- only load files named SKILL.md as Markdown skills (#3175)
- prevent command injection in MacOSTTS (#3129) (#3174)
- avoid ValueError when LLM output has multiple json fences (#3117)
- fix the unsandboxed jinja2 rce problem. (#3111)
- inner\_copy\_and\_install must not report a failed build as success (#3141)
- deserialize table chunks when a separated chunk is present (#3140)
- create\_datasource/create\_flow must POST, not GET (#3138)
- fix CORS allowed origins security (#3123)
- StreamedBytesIO ignores negative read size and drops seek(SEEK\_END) position (#3136)
- handle native bool value in VariablesProvider.\_convert\_to\_value\_type (#3135)
- preserve bucket in simplified fsspec paths (#3134)
- ExcelKnowledge.\_load must handle headerless sheets and multi-sheet workbooks (#3137)
- correct empty/non-list guard in TeiRerankEmbeddings.\_parse\_results (#3133)
- avoid handleChat temporal dead zone (#3132)
- allow Milvus vector store type in TOML configuration (#3127)

## 👥 New Contributors
- @Bartok9
- @Carbene
- @Dellorchid
- @DreamZhongJu
- @Osamaali313
- @XiaoHuo888-hue
- @chuenchen309
- @mumubuku
- @yyyCode

Thank you to all our contributors for making this release possible!
@Aries-ckt, @Bartok9, @Carbene, @Dellorchid, @DreamZhongJu, @Osamaali313, @XiaoHuo888-hue, @chen-alan, @chenliang15405, @chuenchen309, @mumubuku, @yyyCode and 田云钊

中文发版说明: http://docs.dbgpt.cn/zh-CN/docs/next/changelog/Released_V0.8.2
Release Note:http://docs.dbgpt.cn/docs/next/changelog/Released_V0.8.2