MemHub v1.0.6

What’s New / 本次更新

• Memory modes / 记忆模式
  lite (no LLM; keeps raw conversations and searches them directly), pro (LLM extraction) and ultra (adds a knowledge graph and nightly consolidation). Choose one during onboarding or switch any time in Settings.
  lite（不需要大模型，保存原始对话并直接检索原文）、pro（大模型抽取记忆）、ultra（增加知识图谱和每晚自动整理）。首次启动时选择，之后可在设置中随时切换。

• Model Center / 模型中心
  Separate settings for the memory model and the search (embedding) model. Switching the embedding model re-embeds existing memories safely; nothing is lost and you can cancel.
  "记忆整理"和"记忆搜索"模型分开配置。更换嵌入模型时会安全地重新整理已有记忆，不会丢失，可随时取消。

• 11 interface languages / 11 种界面语言
  English, 简体中文, 繁體中文, 日本語, Español, Deutsch, Português, Français, 한국어, Русский, Bahasa Indonesia. Memories are written in the language you choose.
  记忆内容会使用你选择的界面语言。

• Writes retry by themselves / 记忆写入自动重试
  If the model service is misconfigured or down, MemHub pauses, shows a banner and resumes on its own once it works again — no manual retry.
  模型配置有误或服务不可用时，MemHub 会暂停处理并在顶部提示，恢复后自动继续，不需要手动重试。

• Faster and steadier / 更流畅、更稳定
  Fixes write stalls and the App freezing for long periods on machines with many connected apps; live capture now goes only through MemHub's own service.
  修复接入应用较多时写入停滞、App 长时间卡顿的问题；实时记录统一交给 MemHub 后台服务处理。

• More apps and users / 更多应用与多用户
  DeepSeek Harness connects all profiles with one switch; the CLI and API support multiple users (user_id), e.g. one Hermes profile per person.
  DeepSeek Harness 一个开关接入全部 profile；CLI 和 API 支持多用户（user_id），例如每人一个 Hermes profile。

• More reliable upgrades / 升级更可靠
  Fixes the upgrade failures reported in #229, #234 and #240, plus many stability fixes.
  修复 #229、#234、#240 等升级失败问题，以及多项稳定性改进。

Upgrade note / 升级提示： On first launch, MemHub may spend a few minutes checking and organizing your existing memories. Please do not force-quit during this time.
首次启动时，MemHub 可能需要几分钟检查并整理已有记忆，请勿在此期间强制退出。

Built on hy-memory==2.0.14.

Full Changelog / 完整更新记录: v1.0.5...v1.0.6