SuperSonic 0.9.10 released
Updated on:July-14th-2025
Recommendation
SuperSonic 0.9.10 version brings many updates, more powerful performance, and higher usability!
Core content:
1. The new version is supported by 10+ contributors and optimizes high-frequency feedback issues
2. Supports two deployment methods: Zip compression package and Docker image, simplifying the installation process
3. Parsing function upgrade, adding mapping mechanism, integrating rule and LLM parsing paths
Yang Fangxian
Founder of 53AI/Most Valuable Expert of Tencent Cloud (TVP)
0.9.10 has made a lot of optimizations in terms of bugs, usability, and compatibility that users frequently report. Here is a brief overview from the four dimensions of deployment , parsing , permissions , and integration .Docker and installation package deployment have been further simplified to reduce the user's learning cost. Please refer to the article 5 minutes to simply deploy SuperSonic. It is worth mentioning that Docker images already support two architectures: amd and arm:In addition, it is important to note that when upgrading from 0.9.10 to Spring 3.0 (special thanks to mislayming for contributing the PR), the recommended JDK version is at least 21 .Parsing means converting the user's natural language into data query language . In SuperSonic, it includes four stages: schema mapping, semantic parsing, semantic correction, and semantic translation:- Mapping stage: Add a mechanism to map all fields, which can be controlled by parameters in the system settings. If the number of original mapped fields is less than the threshold, all fields will be mapped and LLM will make its own judgment. Note: If the data set has a lot of fields, it is recommended to avoid mapping all fields.
- Parsing phase: The two parsing paths of rule and LLM are integrated. Rule parsing is used as a fallback solution when LLM is unavailable, and as a pre-processing for LLM parsing. In addition, a retry mechanism is added to the parsing process to improve robustness.
- Correction stage: Add an LLM-based corrector to supplement rule correction. It is closed by default and needs to be manually enabled in assistant management.
- Translation phase: The translator was refactored to simplify the abstraction, and the snowflake model was initially supported, and the multi-table join capability was enhanced. In addition, the judgment of the with statement was optimized, especially since a large number of users are still using MySQL 5.7, which does not support with.
In terms of user authentication, SuperSonic has built-in basic account management, and has fixed the password modification and user registration functions in version 0.9.10. If you want to connect with a third-party account system, the system provides two extension points, AuthenticationInterceptor and DefaultUserAdaptor, for rapid customization and development.In terms of user authorization, the much-requested assistant permission management has been added . So far, the four modules of conversation assistant, subject domain, semantic model, and database connection can all be authorized, and the authorization mechanism and operation are consistent.The database has added support for HANA, Starrocks , Presto , Trino , and Kyuubi (special thanks to wwsheng009 and zyclove for contributing PR). These are the mainstream query engines in the industry.The vector library adds support for OpenSearch (special thanks to zyclove for contributing PR). At the same time, the official Docker deployment uses pgvector by default, so that metadata and vector storage reuse the same postgresql instance.