Skip to main content

DesireCore v10.0.91

Release Date: 2026-07-20

v10.0.91 更新概览

New Features​

  • Reasoning Depth Options: Added optional reasoning depth (Fast/Medium/High/Very High/Max) for the GPT-5.6 model series, which can be toggled in the input area to control the depth of thought in responses.
  • Version Gating for Skill Installation: The client version is now automatically validated when installing a skill from the marketplace. Incompatible installations are blocked and an upgrade is prompted, preventing functional issues caused by version mismatches.

Fixes​

  • Fixed issues where continuing generation would rewrite the entire response, old responses persisted after editing and resending, historical branch content could not be properly restored, and messages were duplicated—ensuring coherence in multi-turn conversations.
  • Fixed issues where approval cards remained in a pending state after a service restart, stopping a session incorrectly reported as user rejection, and the approval mode for delegated tasks unexpectedly downgraded to the global policy.
  • Fixed marketplace and installation experience issues, including application installation states not persisting, install button states on skill detail pages not updating, missing local installation entry when no remote device is present, and duplicate display of skill counts.
  • Fixed an issue where background analysis tasks (e.g., topic summaries) repeatedly failed and could not self-heal due to swallowed output limit errors, improving internal processing stability.
  • Fixed a 401 error occurring when the Agent called local APIs via skills in local mode, restoring normal integration with local services.
  • Fixed various UI and interaction defects: multiple dialogs could not be closed with the Esc key, Esc events incorrectly triggered other panels, example email addresses that cannot receive mail were accepted during registration, and the working memory panel lacked user private memories.

Improvements​

  • The application installation and uninstallation processes now display more detailed intermediate states (e.g., Uninstalling, Reinstalling). Services are no longer unexpectedly cleaned up before the operation completes, providing more precise status feedback.
  • Optimized the prompt for the agent management skill: removed the rigid limit on the number of questions, allowing the Agent to control the interaction rhythm autonomously, resulting in more natural conversations.
  • Model configuration updates: expired models are automatically taken offline and can fall back to available models of the same type; support for the GPT-5.6 model series has been added.
  • Automatic topic summarization for official cloud accounts now uses a lower-cost model, reducing computing resource consumption for background tasks.
  • Added system environment descriptions in the runtime environment panel, clearly distinguishing between local tools and hosted runtimes to avoid misunderstandings caused by "not installed" prompts.
  • Enhanced local service security: by default, only necessary internal endpoints are exposed.