Create Your First Agent

中文原版 · ← Back to Overview

Markus’s core is the AI digital worker (Agent) — it proactively accepts tasks, plans and executes autonomously, and delivers acceptable results, growing continuously through persistent memory, tools, and collaboration protocols. This page teaches you how to hire / import your first Agent in the Web UI and have your first conversation.


1. Prerequisites

  • Complete Installation and log in successfully (if you haven’t started yet).
  • Add at least one LLM API Key in the Settings page — otherwise Agents cannot call models.

2. Hire Your First Agent

In the team management area on the Team page, click “Hire Agent” / 聘用 AI 员工:

  1. Specify a Role: built-in roles live in templates/roles/, for example:
    • manager — organization owner (routing and coordination)
    • developer — software engineering developer
    • product-manager — product manager
    • operations — DevOps
  2. Specify a Position: Worker (execution) or Manager (management).
  3. After submitting, the Agent appears in the team card with an online status indicator.

You can also create the team first and then hire: Owner / Admin can create teams on the Team page, set a team Manager, and add members.

Custom Roles (optional)

Create the templates/roles/my-role/ directory and fill in your role definition:

templates/roles/my-role/
├── ROLE.md        # Required: role definition & system prompt
├── SKILLS.md      # Optional: skills & tool permissions
├── HEARTBEAT.md   # Optional: scheduled self-driven tasks
└── POLICIES.md    # Optional: behavior rules

Then assign that role to your Agent.

3. View the Agent Profile

Click an Agent row on the Agents page — the Profile panel opens on the right, showing:

  • Role / Status / Memory / Tools
  • Start / Stop: the two buttons switch depending on status, used to start/stop the Agent (online shows a green dot).
  • Mind tab: view the Agent’s attention state, mailbox queue, and its Notebook (persistent cognitive workspace).

Agents have a three-layer memory (procedural / semantic / episodic); knowledge accumulates across sessions and consolidates automatically — the longer your team runs, the smarter it gets.

4. Talk to Your First Agent

Two ways:

  • Chat page → Agent list: select the Agent for a direct one-on-one chat (conversations are isolated per user, invisible to each other).
  • Team page → click the Agent: opens a one-on-one conversation.

Now give it a concrete task in natural language, for example:

“Help me organize today’s todos, sort them by priority, and list the blockers separately.”

The Agent will accept the task and plan and execute autonomously. You can also @-mention it or ask in a team channel, and it reports its current status in context.

5. Import Skills in Conversation (advanced)

In a conversation, an Agent can load local skill packages directly via discover_tools({ mode: "import" }) — effective immediately, no restart needed. It also supports markus skill import <path> to import community skills from skills.sh / Claude Code, SkillHub / ClawHub, OpenClaw, SOUL.md, AgentScope, MCP servers — plug and play.

6. Next Steps

  • Create Your First Task → — walk through: requirement → task → Kanban execution.
  • To see how multiple Agents divide work and deliver results, see Task Collaboration.
  • For the structure and import of digital assets such as Agents / Teams / Skills, see Assets.