Skip to main content

DesireCore v10.0.68

Release Date: 2026-05-27

v10.0.68 更新概览

New Features​

  • Cloud account login and automatic compute binding: You can now register and log in to DesireCore directly with your email. After logging in, the system will automatically bind your cloud models and free quota, and will automatically sync the available model list going forward.
  • Version announcement popup: After upgrading to a new version, an announcement will pop up once on first launch to help you quickly understand important changes such as the launch of cloud accounts and the retirement of the old public beta compute service.
  • SuperDoc DOCX export: SuperDoc now supports exporting content directly as a DOCX file, which can be opened with common office software such as Word and WPS.
  • Agent automatically compresses the current conversation: When the context gets longer, the Agent can now proactively organize and compress the current conversation to help continue tasks more efficiently.

Fixes​

  • When retrying reply generation, previously generated text and completed tool calls are now preserved, instead of starting over from scratch because of automatic retries.
  • Fixed an issue where long conversations could lose part of the message history after retries, interruptions, unexpected exits, or recovery, making history more complete and reliable.
  • Fixed an issue where image attachments could be lost when retrying messages, making regeneration in image-based conversations more stable.
  • Fixed an issue where conversation records could be missing when exporting an Agent, allowing the original conversation history to be preserved more completely after import.
  • Fixed an issue in DeepSeek reasoning mode where historical reasoning content could be lost, making continuous reasoning conversations more stable.
  • Fixed an issue during cloud model syncing where some Zhipu models could error out or be interrupted, improving availability for conversations using those models.
  • Fixed an issue where the version announcement could fail to display properly in some scenarios because the version number was not injected.

Improvements​

  • When you manually click "Compress Conversation," compression will now be attempted as much as possible even if the conversation does not have many turns, making it more practical for long-text and long-code scenarios.
  • If you are not logged in or a model is unavailable, a clear prompt will now appear before sending a message, and your text and attachment drafts will be preserved to prevent content from being lost unnecessarily.
  • System prompts related to login, insufficient compute, or unavailable models are now smarter and will provide more appropriate guidance based on your login status.
  • The resource panel now includes clearer login guidance and cloud model sync status prompts, making it easier to know when to log in and whether models are being updated.
  • The login popup and account flow have been optimized, including switching between sign-up and login, email verification prompts, password manager support, and the "Remember me" logic, resulting in a smoother overall experience.
  • The old public beta compute entry point has now been fully retired and cleaned up, so the interface and model list are more focused on the officially available cloud account capabilities.