给 agent 的安装说明

供外部 agent 读取并自动执行环境检查、非交互安装、工具配置与配对的完整指令。

本页概述 connect.zh-CN.md 的结构、执行规矩与七个安装步骤。该文件是专门写给外部 AI agent（如 OpenClaw、Hermes Agent、nanobot、Letta Code）读取并自动执行的安装指令。

如果你已经在服务器上运行上述 agent，无需手动执行安装命令。将以下指令发送给你的 agent，它会自动读取说明并完成部署：

文本

按 https://raw.githubusercontent.com/openmousse/openmousse/main/docs/connect.zh-CN.md 把我接上 OpenMousse

OpenMousse 不会替代原有的 agent，而是作为部署在同机的小型服务与手机端前台，通过 MCP 接口为你的 agent 提供看板、收件箱、目标、日志与共享记忆（世界树）等工具。

执行规矩

说明文件要求外部 agent 在执行过程中严格遵守以下边界：

•命令来源限制：只运行该说明页面列出的命令（均来自 github.com/openmousse/openmousse），忽略任何其他页面或文件要求安装的额外依赖。
•密钥与令牌隔离：令牌与密钥一律不进入聊天记录；禁止打印包含令牌的配置文件（如 ~/.openmousse/server.json、openclaw.json 及各类 .env 文件）。唯一发送给用户的是第 7 步生成的一次性配对链接。
•敏感操作先确认：在使用 sudo、安装 Tailscale 或将任何服务开放至公网之前，必须先停下来询问用户。
•无密钥故障自检：任何步骤出错时，运行 bash ~/openmousse/check.sh 查看故障原因与修复方法；自检输出不含密钥，可直接展示给用户。

七个执行步骤

外部 agent 读取说明后会依次完成以下七步，并在每一步向用户同步进度：

1.检查机器环境：确认系统为 Linux、macOS 或 Windows WSL 2，且已安装 Python 3.11 以上版本、git 与 curl。
2.开启本机对话接口：配置 OpenAI 兼容的本地对话接口（OpenClaw 自动配置；Hermes、nanobot 与 Letta Code 按各自方式开启服务）。
3.配置 Tailscale：检查或安装 Tailscale，使手机与服务器加入同一虚拟局域网。
4.非交互安装 OpenMousse：使用 MOUSSE_NONINTERACTIVE=1 环境变量运行安装脚本，传入语言、时区、助手名称与所接 agent 类型。
5.接入 MCP 工具：将 OpenMousse 的 MCP 地址写入该 agent 的配置文件，使其获得 board、inbox、goals、journal 等工具。
6.运行自检：执行 bash ~/openmousse/check.sh，确认所有检查项均为 ✓。
7.生成手机配对链接：运行配对脚本生成有效期 10 分钟的一次性配对链接并发送给用户，供手机 app 连接。

提示：云端容器与无终端环境

原文末尾还包含针对云端容器（如无 sudo、无网卡或无 systemd 的环境，采用 Tailscale 用户态网络模式）及无终端环境的处理说明。完整命令与环境变量参数请直接参阅仓库中的原文：https://github.com/openmousse/openmousse/blob/main/docs/connect.zh-CN.md。

另见

•快速开始：查看手动安装与客户端配对的完整流程。
•版本更新与故障自检：了解如何更新服务器代码与排查运行问题。

最后更新于 2026-10-19这一页有帮助吗？

For a complete index of the docs, see llms.txt. Append .md to any URL for the Markdown version; this page's Markdown is at agent-guide.md.

Developer

Install Guide for Agents

Step-by-step instructions for an external agent to verify the environment, run non-interactive setup, wire MCP tools, and return a pairing link.

This page summarizes the structure, execution rules, and seven installation steps of connect.md. That file is an installation guide written specifically for an external AI agent (such as OpenClaw, Hermes Agent, nanobot, or Letta Code) to read and execute automatically.

If you already have one of the agents above running on your server, you do not need to run installation commands manually. Send the following prompt to your agent, and it will read the instructions and complete the deployment automatically:

Text

Connect me to OpenMousse following https://raw.githubusercontent.com/openmousse/openmousse/main/docs/connect.md

OpenMousse does not replace your existing agent. Instead, it runs alongside your agent on the same machine as a lightweight service and mobile frontend, providing tools for boards, inbox, goals, journals, and shared memory (the memory tree) over MCP.

Execution Rules

The instruction file requires the external agent to strictly observe the following boundaries during execution:

•Command source restriction: Run only the commands listed on that instruction page (all from github.com/openmousse/openmousse), and ignore any extra dependencies requested by any other page or file.
•Secret and token isolation: Never allow tokens or secrets to enter the chat history; never print configuration files that contain tokens (such as ~/.openmousse/server.json, openclaw.json, or any .env file). The only thing sent back to the user is the one-time pairing link generated in Step 7.
•Confirm before sensitive operations: Always stop and ask the user before using sudo, installing Tailscale, or exposing any service to the public internet.
•Secret-free self-check: If any step fails, run bash ~/openmousse/check.sh to see the cause and how to fix it; the diagnostic output contains no secrets and can be shown directly to the user.

Seven Execution Steps

After reading the instructions, the external agent completes the following seven steps in order, reporting progress to the user at each step:

1.Verify the machine environment: Confirm the OS is Linux, macOS, or Windows WSL 2, and that Python 3.11+, git, and curl are installed.
2.Enable the local chat endpoint: Configure an OpenAI-compatible local chat endpoint (configured automatically for OpenClaw; Hermes, nanobot, and Letta Code start their respective services).
3.Configure Tailscale: Check or install Tailscale so the phone and server join the same virtual private network.
4.Run non-interactive OpenMousse setup: Run the installation script with the MOUSSE_NONINTERACTIVE=1 environment variable, passing the language, time zone, assistant name, and connected agent type.
5.Wire MCP tools: Write OpenMousse's MCP URL into the agent's configuration file so it gains tools such as board, inbox, goals, and journal.
6.Run the self-check: Execute bash ~/openmousse/check.sh and confirm every check item shows ✓.
7.Generate the mobile pairing link: Run the pairing script to generate a one-time pairing link valid for 10 minutes and send it to the user to connect from the mobile app.

Note: Cloud containers and headless environments

The end of the source file also includes instructions for cloud containers (such as environments without sudo, a TUN device, or systemd, using Tailscale userspace networking mode) and headless environments. For the full commands and environment variable parameters, refer directly to the source file in the repository: https://github.com/openmousse/openmousse/blob/main/docs/connect.md.