Skip to main content

DesireCore v10.0.122

Release Date: 2026-08-25

v10.0.122 更新概览

New features​

  • Bailian Wanxiang 3.0 video generation: The video generation tool now supports Bailian Wanxiang 3.0 (wan3.0-video / wan3.0-video-prime), including parameters such as first/last frames, reference images/videos/audio, web links, smart duration, audio track, seed, prompt expansion, and watermark.
  • Agent model configuration delegation: Agents can now modify another Agent's model source after approval (such as changing the model or provider); previously these requests were rejected outright.

Fixes​

  • Fixed an issue where a configured provider allowlist would silently disappear after switching between smart routing and a fixed model.
  • Fixed an issue where opening Notification Center would cover the custom-drawn minimize, maximize/restore, and close buttons on Windows/Linux.
  • Fixed an issue where new messages could be written to a session pending deletion when an Agent cleared its conversation history while a run was still in progress; the pending clear is now resumed after a restart, and a new conversation can begin normally once clearing is complete.
  • Fixed multiple issues with saving heartbeat preferences: system defaults no longer pollute user settings, changing a single toggle no longer unexpectedly restores muted alerts, switching Agents no longer writes settings across Agents, and settings are reliably saved before the panel is closed.
  • Fixed an issue where the Solver connector's "read → modify → save" flow was rejected as a whole because of extra fields.
  • Fixed an issue where cloning an Agent that used a fixed model would unexpectedly rewrite its model configuration to use smart routing.
  • Fixed an issue where the Agent settings dialog was visible but clicks passed through to the underlying profile page.
  • Fixed a security vulnerability where URL-encoded variants could bypass authentication for the governance API, and prevented misleading copy from being injected into approval summaries by the model.

Improvements​

  • Removed the "Confirm human governance action" dialog from governance operations. The system now automatically recognizes actions from trusted interfaces, reducing daily interruptions.
  • Improved dialog focus management: the topmost dialog now exclusively owns focus cycling and Esc-to-close, while lower dialogs are automatically deactivated to reduce accidental actions.
  • LLM routing configuration read/write logic has been fully consolidated, making behavior among model switching, allowlists, and fixed models more consistent and stable.