跳到正文Skip to content
OpenMousse
完整的文档索引见llms.txt。在网址后加.md可以拿到 Markdown 版本；本页的 Markdown 版本是agents.zh-CN.md。
功能

Agents

了解主对话、功能 Agent、会议纪要、群聊、赛博与管家的分工和边界。

在 OpenMousse app 的对话列表中即可查看并进入所有 Agent；列表第一行固定为主对话，其余为各功能 Agent、赛博与群聊。

OpenMousse 将日常事务、长期思考与对外沟通分给不同角色的 Agent 处理。你可以在主对话中交代需求，由主对话转交给对应的功能 Agent 或召集会议；也可以直接点进任意一个功能 Agent 的对话、看板与独立记忆页面与它沟通。

各类角色与分工

  • •主对话：系统的总入口。负责接收你的日常交办、读写世界树、向功能 Agent 分派任务，以及发起并主持多 Agent 会议。每晚日结后，主对话也可能把你反复处理的事务提议为新的功能 Agent，生成方案卡放入收件箱等你确认。
  • •功能 Agent：每个功能 Agent 只负责一类事，其中代办 Agent、学习与办公 Agent 默认启用，健身 Agent、饮食 Agent、睡眠 Agent 以及其他领域的 Agent（例如研究 Agent）可按需新建。每个功能 Agent 拥有独立的对话、记忆、工具与看板；看板集合支持五种视图：表格、看板、日历、时间线、图表。
  • •赛博 即将推出：拥有独立的性格、记忆与模型，在特定方向上长期辅助你思考（例如一位追问到底的质疑者或一位学习教练）。它的资料页是一张证件，仅包含身份、记忆、装载三项。赛博不直接访问你的数据，查资料或改动数据均交由其「装载的功能 Agent」处理：只读结果直接返回，修改数据则生成确认卡，要写进世界树也须先生成确认卡（或进入「待确认」）经你确认。
  • •管家 即将推出：对外代表你，按亲近、朋友、同学三档关系回答好友及好友 agent 的询问（例如核对周四晚上是否有空）。管家仅读取你按该档位开放的日程、近况、分享过的内容、学习笔记或住址，看不到世界树与健康数据。涉及时间、金钱、承诺的事项，管家一律回答「这个得（你的名字）本人定，我去问一下。」并生成确认卡交给你；每一句对外回复在发出前均经 Doorman 复查。

会议、纪要与群聊

  • •会议与纪要：遇到需要跨领域协同的事，可由主对话发起并主持短会，你也可以在 app 内自己发起。会上几个 Agent 轮流发言，每轮每个 Agent 最多发言一次，并可 @ 其他成员；会上只讨论，不改动任何数据。一轮结束后，主对话出具一张会议纪要卡（包含决定与待办，每条待办标明负责的 Agent 与时间），出现在主对话中。经你确认后，待办才会写入日程与对应 Agent 的看板；你也可以补一句开启下一轮。
  • •群聊：几个 Agent（包括赛博）在一个群内讨论，你 @ 谁谁回答，不生成会议纪要。会议结束后也可以保留为群聊。

不在这里的东西

  • •没有日程 Agent 与朋友 Agent：日程由主对话与各功能 Agent 共用并写入今天页；好友互动与对外回复由「好友」功能及「管家」负责。
  • •没有跨用户的 Agent 模板库：功能 Agent 均在你的服务器上通过主对话方案卡或 app 内表单按需建立。
  • •管家不理会陌生人，也不自决承诺：只有互为好友的联系人能向管家提问；管家回复使用不带任何工具的模型调用，不能自行答应约会或索取请求。

另见

  • •建立并管理功能 Agent：了解通过主对话方案卡或 app 表单新建功能 Agent 的步骤。
  • •世界树：了解各角色读写长期记忆的权限边界。
  • •先问你（收件箱）：了解确认卡的生成与审批机制。
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 agents.md.
Features

Agents

Understand the roles and boundaries of main chat, functional agents, meeting minutes, group chats, Cybers, and the Butler.

You can view and open all agents from the chat list in the OpenMousse app; the first row is always main chat, followed by your functional agents, Cybers, and group chats.

OpenMousse divides everyday tasks, long-term thinking, and external communication across agents with distinct roles. You can state a request in main chat and let main chat hand it off to the appropriate functional agent or call a meeting, or you can open any functional agent's chat, board, and dedicated memory directly.

Roles and responsibilities

  • •Main chat: The primary entry point to the system. It receives your everyday requests, reads and writes the memory tree, dispatches tasks to functional agents, and calls and moderates multi-agent meetings. After the nightly settlement, main chat may also propose turning a recurring task into a new functional agent by placing a proposal card in your inbox for confirmation.
  • •Functional agents: Each functional agent handles a single domain. The Errands agent and the Study & Work agent are enabled by default; fitness, diet, sleep, and other domain agents (such as a research agent) can be created as needed. Each functional agent has its own chat, memory, tools, and board; board collections support five views: Table, Kanban, Calendar, Timeline, and Chart.
  • •Cybers Coming soon: Long-term thinking partners with their own persona, memory, and model for a specific direction (such as a relentless questioner or a study coach). A Cyber's profile page is an ID card containing only three sections: identity, memory, and mounted agents. A Cyber does not access your data directly; any data lookup or modification is handled by its mounted functional agents: read-only results are returned directly, data modifications generate an approval card, and writing to the memory tree also requires generating an approval card (or sending the entry to Needs your approval) for your confirmation.
  • •The Butler Coming soon: Represents you externally, answering inquiries from friends and your friends' agents according to three relationship tiers: close, friend, and classmate (for example, checking whether you are free Thursday evening). The Butler reads only the schedule, recent status, shared items, study notes, or address that you open to that tier, and has no access to the memory tree or health data. For anything involving time, money, or commitments, the Butler always replies, "That's up to (your name) personally—let me ask." and generates an approval card for you; every outbound reply is reviewed by Doorman before it is sent.

Meetings, minutes, and group chats

  • •Meetings and minutes: When a matter requires coordination across domains, main chat can call and moderate a short meeting, or you can start one yourself in the app. Participating agents speak in turns, at most once per round, and can @ other members; during the meeting they only discuss and do not modify any data. At the end of a round, main chat produces a meeting minutes card (containing decisions and action items, with each action item specifying the responsible agent and time) directly in main chat. Action items are written to your schedule and the respective agents' boards only after you confirm; you can also add a note to start the next round.
  • •Group chats: Several agents (including Cybers) discuss in a single group where whichever agent you @ replies, without generating meeting minutes. A meeting can also be kept as a group chat after it ends.

What stays out

  • •No Schedule agent or Friends agent: Your schedule is shared across main chat and functional agents and displayed in Today; friend interactions and external replies are handled by Friends and the Butler.
  • •No cross-user agent template store: Functional agents are created on demand on your own server via a proposal card in main chat or an in-app form.
  • •The Butler ignores strangers and makes no commitments on its own: Only mutual friends can ask the Butler questions; Butler replies use a tool-free model call and cannot accept appointments or requests on your behalf.

See also

  • •Create and Manage Functional Agents: Learn the steps for creating a functional agent via a proposal card in main chat or the in-app form.
  • •Memory Tree: Learn the permission boundaries for reading and writing long-term memory across roles.
  • •Approvals (Inbox): Learn how approval cards are generated and reviewed.
Last updated 2026-10-19Was this page helpful?

这一页给 agent 读的版本：在网址后加 .md 就是下面这份纯文本，和给人看的那页由同一份内容生成。Agent-readable version of this page: append .md to the URL to get the plain text below, generated from the same source as the human view.