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create-agent.zh-CN.md。
指南

建立并管理功能 Agent

在主对话中通过方案卡或在 app 内新建功能 Agent，并为它配置职责与模型。

完成本指南后，你将建立一个新的功能 Agent（例如专门整理行业报告的研究 Agent），为它设定职责与所用模型，并了解如何通过直接对话、主对话转交或召集会议与它协作。本指南假设你已部署 OpenMousse 并完成客户端配对。

前提条件

  • •已运行 OpenMousse 服务器并配对手机或网页客户端。
  • •服务器已配置至少一个可用的模型服务（或已接入 OpenClaw / Hermes）。

隐私说明

  • •记忆与数据隔离：每个功能 Agent 拥有独立的对话记录、独立记忆与专属看板，与其他功能 Agent 分开存放。
  • •共享世界树：各功能 Agent 在回答前均可查询世界树，并在得知关于你的新事实、偏好、决定或结论时写入世界树（每天 03:45 日结时也会汇总写入）。
  • •外部应用不可见：你手动授权接入世界树的外部 AI 应用（如 Claude、ChatGPT）无法读取任何功能 Agent 的独立记忆、对话记录与看板数据。

步骤

1. 选择一种方式建立功能 Agent

代办和「学习与办公」两个 Agent 默认启用；针对健身、饮食、睡眠等其余日常领域，你可以按以下任意一种方式按需建立对应的功能 Agent（从一句话描述到试聊满意再创建的引导式新建流程 即将推出）：

建立方式操作步骤
方式 A：在主对话中通过方案卡建立1. 打开对话列表第一行的「主对话」。2. 说明你希望新 Agent 负责的事务，例如：「帮我建一个研究 Agent，专门整理行业报告和纪要。」3. 主对话会生成一张包含名称与职责设定的方案卡；核对无误后点击确认，系统即建立该 Agent。
方式 B：在 app 内手动填写建立1. 在 app 的「对话」页顶部点金色 + 按钮（「新建」）→「新建 Agent」（若还不确定，也可在该页面选择「尚未确定？先讨论方案」）。2. 依次填写该 Agent 的名称、职责说明并选择它所使用的模型，保存后即完成创建。
方式 C：批准日结后的收件箱提议每晚 03:45 日结后，若主对话发现某类事务被你反复处理，会自动提议将其独立为一个新的功能 Agent，并生成确认卡放入收件箱与今天页的「待处理」；你点击同意后即建立该 Agent。

2. 与功能 Agent 日常协作

建立完成后，你可以通过三种路径使用该功能 Agent：

  1. 1.直接进入独立对话：在对话列表中点开该功能 Agent，直接向它交代任务或查阅它的独立记忆。
  2. 2.由主对话转交：在主对话中直接说出需求，主对话会识别事务类型并转交给对应的功能 Agent 处理。
  3. 3.召集多个 Agent 开短会：当一项决定涉及多个领域（例如出差期间如何同时调整研究进度、睡眠与饮食），你可以在主对话中提出（或在 app 内直接发起会议）。主对话会召集相关功能 Agent 轮流发言（每轮每个 Agent 最多发言一次，可互相 @），讨论结束后在主对话中生成一张会议纪要卡，列出决定与每条待办的负责 Agent 及时间。
注意：会议纪要须经你确认才写入数据

会议过程中各 Agent 仅作讨论，不会修改任何日程或看板数据。会议纪要卡出现在主对话中（不在今天页的「待处理」里）；只有当你点击确认后，纪要中的待办事项才会正式写入你的日程以及对应功能 Agent 的看板。你也可以补充意见并开启下一轮讨论。

3. 切换看板视图管理结构化数据

每个功能 Agent 均配有专属看板，用于维护它所负责领域的结构化记录（如文献台账、训练计划或待办事项）。进入该 Agent 的看板后，你可以按需要在五种视图之间切换：表格、看板、日历、时间线与图表。

成功的标志

当以下各项均成立时，说明新的功能 Agent 已投入工作：

下一步

  • •Agents：全面了解主对话、功能 Agent、群聊、赛博与管家的权限边界。
  • •先问你（收件箱）：了解功能 Agent 执行操作时的三档权限与确认卡机制。
  • •今天页与 04:00 日结：了解各功能 Agent 如何向今天页投递报告并在凌晨完成日结。
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 create-agent.md.
Guides

Create and Manage Functional Agents

Create functional agents via proposal cards in main chat or directly in the app, and configure their roles and models.

After completing this guide, you will have created a new functional agent (for example, a research agent dedicated to organizing industry reports), configured its role and model, and learned how to work with it through direct chat, main chat handoffs, or multi-agent meetings. This guide assumes you have deployed OpenMousse and paired a client.

Prerequisites

  • •A running OpenMousse server paired with a phone or web client.
  • •At least one model provider configured on the server (or a connected OpenClaw / Hermes instance).

Privacy

  • •Isolated memory and data: Each functional agent has its own chat history, dedicated memory, and dedicated board, stored separately from other functional agents.
  • •Shared memory tree: Every functional agent can query the memory tree before replying and write new facts, preferences, decisions, or conclusions about you to the memory tree as it learns them (and in summary during the daily 03:45 settlement).
  • •Invisible to external apps: External AI apps that you authorize to access the memory tree (such as Claude or ChatGPT) cannot read any functional agent's dedicated memory, chat history, or board data.

Steps

1. Choose a way to create a functional agent

The Errands and Study & Work agents are enabled by default. For other everyday areas such as fitness, diet, or sleep, you can create a functional agent on demand using any of the following methods (a guided creation flow from a one-line description to a test chat before creating Coming soon):

MethodSteps
Method A: Via a proposal card in main chat1. Open Main chat in the first row of the chat list.2. Describe what you want the new agent to handle, for example: "Create a research agent to organize industry reports and meeting notes."3. Main chat generates a proposal card with the agent's name and role definition; review it and tap confirm to create the agent.
Method B: Manually in the app1. Tap the gold + button (New) at the top of Chats → New Agent (or choose Not sure yet? Talk it through on that screen).2. Enter the agent's name and role description, select its model, and save to finish creating it.
Method C: Approve a post-settlement inbox proposalAfter the nightly 03:45 settlement, if main chat notices a recurring type of task you handle repeatedly, it automatically proposes spinning it off into a new functional agent and places an approval card in your inbox and To do in Today; tap approve to create the agent.

2. Collaborate with a functional agent day to day

Once created, you can work with a functional agent in three ways:

  1. 1.Chat with it directly: Open the functional agent from the chat list to assign tasks directly or inspect its dedicated memory.
  2. 2.Hand off from main chat: State your request in main chat; main chat identifies the task type and hands it off to the corresponding functional agent.
  3. 3.Call a short multi-agent meeting: When a decision spans multiple areas (for example, adjusting research progress, sleep, and diet during a business trip), raise it in main chat (or start a meeting directly in the app). Main chat calls the relevant functional agents to speak in turns (each agent speaks at most once per round and can @ other participants), then generates a meeting minutes card in main chat listing the decisions and the responsible agent and time for each action item.
Important: Meeting minutes only write data after you confirm

During a meeting, agents only discuss and do not modify any schedule or board data. The meeting minutes card appears in main chat (not under To do in Today); action items from the minutes are written to your schedule and the respective functional agents' boards only after you tap confirm. You can also add a comment to start another round of discussion.

3. Switch board views to manage structured data

Every functional agent comes with a dedicated board for maintaining structured records in its domain (such as a literature log, training plan, or task list). Inside an agent's board, you can switch among five views as needed: Table, Kanban, Calendar, Timeline, and Chart.

Verification

Your new functional agent is ready for use when all of the following are true:

Next steps

  • •Agents: Understand the permission boundaries across main chat, functional agents, group chats, Cybers, and the Butler.
  • •Approvals (Inbox): Learn about the three-tier permission model for tool calls and how approval cards work.
  • •Today View and 04:00 Settlement: Learn how functional agents deliver reports to Today and complete the nightly settlement.
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.