Skip to content

RELEASE_NOTES.md

Latest commit

 

History

History
44 lines (24 loc) · 3.03 KB

RELEASE_NOTES.md

File metadata and controls

44 lines (24 loc) · 3.03 KB

Scalytics Copilot - Version 4

This release introduces significant architectural updates focused on enhancing the Deep Search tool's capabilities and providing greater transparency into agent operations. Key changes include the integration of multilingual support, a new agent reasoning log, and a bot trap detection system for more robust web scraping. The update also includes performance improvements, bug fixes, and dependency updates.

Major Features

  • Multilingual Support for Deep Search (#135): The Deep Search functionality has been upgraded to support multiple languages. The system now includes language detection and translation capabilities, allowing it to process queries and sources in various languages.

  • Agent Reasoning Transparency (#128): A persistent reasoning log has been implemented for the Deep Search agent. This provides a detailed, step-by-step record of the agent's decision-making process, including data retrieval, analysis, and synthesis steps. The logs are available for both real-time and historical review, increasing the transparency of the agent's operations.

New Features

  • Bot Trap Detection and Timeout Management (#130): A new system has been implemented to detect and manage unresponsive or malicious domains during web scraping. This feature improves the reliability of data collection by temporarily blacklisting problematic domains.

  • Chat History Summarization (#129): A summarization service has been added to manage long-running conversations. When a chat history exceeds the context window, the service summarizes the conversation to maintain context.

Improvements

  • GPU Statistics: The vLLM service has been enhanced to provide more detailed GPU statistics, including a list of loaded models and the tensor parallel size.

  • Model Activation (#131): The model activation process has been refactored to improve error handling and provide clearer user feedback.

  • Citation Handling (#127): The citation generation process in report synthesis has been improved with standardized citation markers and better URL normalization.

  • Reasoning Log Display (#128): The user interface for the reasoning log has been redesigned for improved clarity and usability.

Bug Fixes

  • Language Support in Synthesis (#135): The processing_language parameter has been added to the synthesis method to ensure correct handling of different languages.

  • Hardware Information Endpoint (#132): The getHardwareInfo function has been simplified by removing redundant code.

  • Chat Title Editing (#133): The functionality to edit chat titles has been removed.

  • Reasoning Log Display (#128): Issues with task_id propagation and conditional rendering in the reasoning log display have been resolved.

  • LLM Timeout (#130): The LLM response timeout has been increased to 300 seconds.

Dependency Updates

  • axios has been updated to version 1.12.0.
  • mermaid has been updated to version 11.11.0.
  • brace-expansion has been updated to version 1.1.12.
  • form-data has been updated.