Skip to main content

DesireCore v10.0.14

Release Date: 2026-03-19

v10.0.14 Release Overview

New Features​

  • Message Read Aloud: You can now have agent messages read aloud with a single click, with support for toggling between play and stop. On desktop, the system will prioritize the native system voice and automatically fall back to the browser voice when unavailable.
  • /new Context Isolation: Entering /new now starts a truly independent new conversation, with a clear “context isolated” divider shown in the chat.
  • New Model Support: The beta provider now includes GLM-5-Turbo and MiniMax-M2.7. At the same time, the built-in MiniMax provider presets have also been updated to include MiniMax-M2.7-related models for easier direct selection.

Fixes​

  • Fixed an issue where the agent could still carry over context from the previous conversation after /new. New conversations will no longer “remember” what was just discussed.
  • Fixed an issue where switching models could still result in the old model being used. Conversations will now reliably use the currently selected model after switching.
  • Fixed an occasional error when sending messages that could prevent the conversation from continuing normally.
  • Fixed an issue where models with small context windows could easily trigger 400 errors when reading large files, and added clearer Chinese prompts when the context limit is exceeded.
  • Fixed an issue where models such as MiniMax could continuously return 400 errors in some gateway scenarios due to incomplete tool parameters.
  • Fixed issues on Windows where dragging and dropping files would show the full file path in messages, and where the agent could occasionally fail to locate files.
  • Fixed an issue where manually adding an agent in a local directory would not automatically refresh the agent/session list in the app.
  • Fixed an issue where creating or editing agent configurations could fail validation and prevent saving due to inconsistencies between $schema and the version number format.

Improvements​

  • File reading and context trimming strategies are now smarter. Models with small context windows will be more stable when handling long text and Chinese content, with fewer unnecessary retries.