Starstruk Logo MarkStarstruk

v0.8.1

mherrmann/heliumv0.8.1Jun 18, 2026by github-actions[bot]

AI Summary

This release introduces vector search support for Qdrant and Valkey, new data source integrations for StarRocks and openGauss, and LiteLLM as an embedded AI gateway. It also includes significant security hardening, bug fixes, and enhancements to context management and task tracking.

Key Highlights

  • Support Schedule task and MCP connector
  • add Valkey vector store integration
  • add LiteLLM as an embedded AI gateway provider
  • Implement VARBINARY and BINARY types for StarRocks
  • Harden file permissions for config files

New Features

  • VARBINARY and BINARY types for StarRocks
  • Valkey cache and vector store integration
  • LiteLLM AI gateway provider
  • Context management and task-plan tracking UI
  • openGauss datasource support

Full Release Notes

## ✨ Enhancements

- Support Schedule task and MCP connector (#3095)
- add CLIENT SETNAME to Valkey vector store client (#3090)
- upgrade MiniMax default model to M3 (#3093)
- Harden ~/.dbgpt/configs/\<profile>.toml file permissions to 0o600 (#3077)
- adds context management, task-plan tracking, and the corresponding frontend UI. (#3053)
- Implement VARBINARY and BINARY types for StarRocks (#3062)
- add Valkey cache integration (#3057)
- make distance metric configurable (#3044)
- add Valkey vector store integration (#3051)
- add LiteLLM as an embedded AI gateway provider (#3043)
- Adds support for vector search with Qdrant (#3034)
- add openGauss datasource support (#3007)
- cache ConnectorManager.get\_connector with TTL (~63s -> \<10ms on warm cache) (#3046)

## 🐞 Bug fixes

- require local runtime opt-in (#3092)
- support DeepSeek V4 Pro (#3079)
- use size chunking by default for markdown knowledge (Fixes #3030) (#3033)
- restrict personal skill script execution (#3071)
- validate example file names (#3066)
- validate skill upload filenames (#3065)
- handle knowledge space id responses (#3070)
- constrain python upload filenames (#3064)
- tolerate multi-step ReAct output (#3074)
- Implement VARBINARY and BINARY types for StarRocks (#3062)
- tell LLM the table list is a TOP-K subset, not the full DB (#3045)
- fix the insufficient length of the content column in the gpts\_messages table (#3055)
- make MilvusStore compatible with milvus 2.5+ (#3042)
- per-chunk error tolerance + per-db indexing lock (#3040)
- override get\_fields() for MSSQL with SQL Server-correct INFORMATION\_SCHEMA (#3039)
- add authentication dependency to knowledge module endpoints (#3038)
- BranchOperator incorrectly skips shared downstream nodes (fixes #2935) (#3035)
- honor configured Tongyi embedding model name (Fixes #3029) (#3032)
- use explicit UTF-8 encoding when writing code\_interpreter temp script (#3023)
- Fix DuckDB datasource creation from Web UI (#3009)


## 👥 New Contributors

- @Aalron
- @Anush008
- @Artoriasn
- @Goodnight77
- @JAE0Y2N
- @RheagalFire
- @atao2004
- @daric93
- @github0070070
- @hankchrome
- @he-yufeng
- @mryanzhicong
- @wolfkill


Thank you to all our contributors for making this release possible!
@Aalron, @Anush008, @Aries-ckt, @Artoriasn, @Goodnight77, @JAE0Y2N, @RheagalFire, @atao2004, @chenliang15405, @daric93, @github0070070, @hankchrome, @he-yufeng, @honglei, @mryanzhicong, @octo-patch and @wolfkill


中文发版说明： https://www.yuque.com/eosphoros/dbgpt-docs/cehdiidrbm8kw1md
Release Note：http://docs.dbgpt.cn/docs/next/changelog/Released_V0.8.1