Loci

io.github.AskLoci由 GitHub 账号 AskLoci 发布v0.1.7更新于 Oct 30, 2026

Run local models on your Apple silicon Mac with Loci. Requires macOS 26.2 or later.

已验证STDIO仅桌面文件与存储AI 与机器学习

概览

AI 生成的概览

Loci 让 AI 助手调用运行在 Apple 芯片 Mac 上的本地模型,完成摘要、改写、分类、结构化输出和批处理任务。

可以做什么
  • 调用本地模型进行摘要、改写、分类、结构化输出和批处理
  • 处理文件并从文档中提取内容(取决于已安装的 Loci 应用版本)
  • 在应用版本支持时搜索文件夹并转写音频
  • 检查与 Loci 应用的连接,并协助选择已安装或可下载的模型
  • 在开始模型工作前修复失效或被拒绝的连接
适用场景
适合希望把摘要、改写、分类或批量文本处理交给本机运行的模型而非云端服务,或让助手读取并总结你授权的文件的场景。
运行要求
需要 Apple 芯片 Mac、macOS 26.2 或更高版本,并安装且运行 Loci 直装应用,同时在应用的高级设置中启用 Agent Access。客户端通过插件市场或 MCPB 包安装;Claude Code 与 Gemini CLI 方式需要 Node.js 22 或更高版本,MCPB 包则无需 Node.js 或 Python 运行时。可用工具取决于应用版本。
安装前请注意
文件工具可读取你授权的任意路径,摘要、提取值、引文和搜索片段可能泄露这些文件中的信息。对话中共享的提示、路径和结果遵循该助手自身的数据政策。连接器读取 Loci 仅所有者可访问的连接文件,并仅通过回环地址发送凭据;应用与插件分别更新,各客户端版本可能不同。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Loci,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

Loci for agents

Use models running on your Mac for summaries, rewriting, classification, structured output and batch processing. Compatible Loci app versions also provide file processing, document extraction, folder search and audio transcription.

Requires an Apple silicon Mac, macOS 26.2 or later, and the Loci direct-download app. Available tools depend on your app version.

Codex

Add the Loci marketplace:

sh
codex plugin marketplace add AskLoci/loci-plugins

In the desktop app, open the Plugins Directory, select Loci, and install the Loci plugin. If the marketplace does not appear, restart the desktop app. Start a new chat and ask: “Help me set up Loci.”

Download the Codex plugin (0.1.8).

Claude Code

Requires Node.js 22 or later.

Add the marketplace and install Loci:

sh
claude plugin marketplace add AskLoci/loci-pluginsclaude plugin install loci@askloci

Start a new conversation and ask: “Help me set up Loci.” Setup and data handling.

Cursor

Open Customize → Plugins and add https://github.com/AskLoci/loci-plugins as a repository source. Install Loci, start a new conversation, and ask: “Help me set up Loci.” Setup and data handling.

Gemini CLI

With Node.js 22 or later on your PATH, install the extension:

sh
gemini extensions install https://github.com/AskLoci/loci-plugins --ref main

Start a new Gemini CLI session on your Mac and ask “Help me set up Loci.” Run /mcp to see which Loci tools are available. If you already configured a manual server named loci, Gemini uses it first. To switch to the extension, remove that entry with gemini mcp remove loci --scope user or --scope project, matching where you added it, then restart Gemini CLI.

Antigravity

Set up Loci in Antigravity.

Hermes Agent

Requires Hermes Agent 0.23.0 or later running on the same Mac as Loci.

sh
hermes plugins install https://github.com/AskLoci/loci-plugins/tree/main/loci-hermes --no-enablehermes plugins enable loci

Ask “Help me set up Loci.” Setup and data handling.

Other MCP clients

Use the Loci MCP bundle with a local macOS client that supports MCPB, or configure the signed connector as a local stdio server. No Node.js or Python runtime is required.

Download packages

Connect Loci

Open Loci and enable Settings → Advanced → Agent Access. Keep Loci running while using the plugin. The setup tool checks the connection and helps choose an installed model or a compatible download. Existing models are shared with the app.

If the saved connection is stale or refused, setup gives a connection-repair step before any model work. App and plugin updates are separate; each client can have a different plugin version.

Data and file access

Model inference runs on your Mac. Prompts, paths and returned results shared in your assistant conversation follow that assistant’s data policies. File tools can read paths you authorize; summaries, extracted values, quotes and search passages can reveal information from those files.

The connector reads Loci’s owner-only connection file and sends its credential only over loopback. It has no separate analytics or external reporting endpoint. Loci’s data practices are described in its privacy policy.

Support · Privacy · Terms

来源:README.md,提交 ae9ada8

工具

0
SourceWeft 曾在隔离沙箱中以无凭据方式启动该软件包一次,以列出其工具,但未能完成:该软件包类型无法在沙箱中运行。

版本历史

2
  1. v0.1.7最新Oct 30, 2026
  2. v0.1.3Oct 29, 2026

评分

已满足适用检查项的 40%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

    请在仓库或注册中心条目中声明许可证。