Skip to content

Releases: justinjia0813/the-last-brain

0.2.2 — 作者署名与产品介绍

Choose a tag to compare

@justinjia0813 justinjia0813 released this 15 Sep 07:46

0.2.2 — 作者署名与产品介绍

  • 插件作者统一显示为 justinjia0813,并链接到作者的 GitHub 主页。
  • 重新整理仓库首页,加入深浅色实机演示图,以及独立的安装、使用和隐私说明。
  • 保留 0.2.1 的全库 Markdown 检索、来源引用、只读边界和对话记忆功能。

升级只替换 main.js、manifest.json、styles.css,保留已有对话、记忆和模型配置。

0.2.1 · 全库检索与界面还原

Choose a tag to compare

@justinjia0813 justinjia0813 released this 15 Sep 07:21

0.2.1 — 全库检索与界面还原

  • 移除每次读取 20 MB、单篇 1 MB 的截断,逐篇检索全部未排除且可读取的 Markdown 笔记。
  • 复用 Obsidian 读取缓存,只保留排名靠前的相关片段;扫描期间显示进度并支持停止。
  • 分别显示已检索、主动排除和读取失败数量,避免将大小限制与读取错误混为一谈。
  • 聊天默认在主标签页打开;恢复已确认原型的品牌侧栏、消息气泡、字体层级、深浅主题和窄窗口布局。
  • 来源改为带编号跳转的抽屉,模型配置提供预设地址、模型候选与真实连接测试。
  • 保留现有聊天记录、来源快照、记忆、配置及只读限制。

全库覆盖指本地 Markdown 检索范围,不包含附件解析,也不表示把整库发送给模型。发送的仍是受数量和字符预算限制的相关片段。当前采用关键词检索,不能保证同义词或语义召回。

核心与原生界面检查通过;实库验证覆盖全部 4,832 篇 Markdown,读取失败为 0。连接测试仅访问本机模拟服务,未调用真实付费模型。移动端尚未实测。

升级只替换 main.js、manifest.json、styles.css,保留原有 data.json。

0.2.0 · 自适应聊天与模型配置

Choose a tag to compare

@justinjia0813 justinjia0813 released this 15 Sep 06:10

0.2.0 — 自适应聊天与更方便的模型配置

  • 重做聊天工作区:宽面板显示对话历史,窄面板可折叠历史;对话内容独立滚动,输入框始终可用。
  • 跟随 Obsidian 深色和浅色主题。
  • 新增模型配置窗口,支持常用服务地址预填、密钥显示切换、输入校验和真实连接测试。
  • 记忆草稿可编辑后保存或确认,确认后才用于后续问答。
  • 支持停止等待并丢弃迟到回答;服务端可能仍在处理或计费。
  • 保留 0.1.0 的对话、记忆、配置及只读限制。

连接测试仅发送固定测试问候,不发送笔记;可能产生所选服务的调用费用。聊天仍以纯文本整段返回。Markdown 笔记检索范围、读取上限不变。移动端和真实付费模型未在本次发布中验证。

升级时只替换 main.js、manifest.json 和 styles.css,请保留原有 data.json。

0.1.0

Choose a tag to compare

@justinjia0813 justinjia0813 released this 15 Sep 02:06

The Last Brain 0.1.0

First public test release: read-only vault conversations, local chat history, source excerpts, and reviewed memories.

Install

Download the-last-brain-0.1.0.zip below. Extract the the-last-brain folder into your vault's .obsidian/plugins/, then enable The Last Brain in Obsidian's community plugin settings. Requires Obsidian 1.11.4+.

The individual main.js, manifest.json, and styles.css assets are also available for manual or community-directory installation.

Setup

Provide an OpenAI-compatible endpoint, model name, and key in the settings. Save the key, enable sending context, then open the chat ribbon icon. The interface is currently in Chinese.

What to test

  • Ask about a specific topic from your Markdown notes and inspect the cited excerpts.
  • Reload the plugin and check that conversation history remains available.
  • Generate a memory draft, review it, confirm it, and ask a related question in a new conversation.

Only plugin-owned history and memory data are written. Notes and attachments are read-only. Selected context is sent to your chosen model provider only on your action; provider fees may apply.

Validation and limits

Ten automated checks, TypeScript checks, and the build pass. Native macOS Obsidian tests with a synthetic local model covered chat, source navigation, exclusions, memory reuse, failure/retry, persistence, and deletion. No note writes occurred. Real provider response quality and mobile behavior have not been tested. Retrieval currently uses Markdown keyword matching, not semantic search.

This GitHub release is ready for testing; it is not yet approved or listed in the Obsidian community directory.