2025.10.31
Smart Vision Ver4.4 have been released, with the following key upgrades:
- Application
Added Prompt Compare and Debug: Agent Application and Writing Application support entering two versions of prompts to compare LLM responses under different prompts, improving prompt-optimization efficiency.

Added Model Compare and Debug: Agent Application, Data Application and Writing Application support comparing the response differences of multiple models under the same prompt to facilitate model evaluation and optimization.

Writing Application Supports mounting the Enterprise Data Warehouse,enabling automatic generation of business analysis reports with charts, achieving an end-to-end workflow from data to report.

Data Application Enhancements: Added Tracing, allowing click-through navigation on fields/terms/data warehouses within orchestration to accelerate issue locating and analysis optimization.

Optimized AI Application Orchestration Experience: Updated model-selection entry for more intuitive operation; conversation starter support rich-text editing; preset questions and command support drag-and-drop sorting for flexible dialog-flow adjustments.


Workflow Upgrades: ① Added workflow duplication for one-click copying; ② Resolved collaborative editing conflicts—system will notify and sync the latest version.

- Data and Capability
Enterprise Knowledge Base: Knowledge governance enhancements a. Added Meta Information: Provides essential contextual information for text chunks, enriching chunk content and improving knowledge-retrieval accuracy.

b. Enhanced Multimodal Recognition: Supports image-content recognition and outputs structured parsing results.

Enterprise Data Expansion: Added support for mainstream databases/data warehouses such as Hive, Doris, Amazon Redshift, and Snowflake, covering more diverse data assets.

Added MCP Deploy Logs: Added deploy Logs capabilities to quickly identify deployment issues and improve troubleshooting efficiency.

Optimized Document Review Tool: Simplified document-review template configuration, focusing on review criteria and element extraction; key information is extracted before review to improve accuracy and issue localization.

- Super Entrance
Added Data Analytics: Supports uploading custom data files and building semantic data models through manual modeling or AI modeling, enabling natural-language data querying and analysis.

Added Deep Research: Supports one-click initiation of research tasks. Through a visual workflow (“Start Analysis → Preliminary Research → In-Depth Research → Report Writing”), the system generates structured research reports and supports feedback on report quality for continuous improvement.
