mycel
Tutorial: Create Your First Agent
DocsGetting StartedTutorial: Create Your First Agent

Tutorial: Create Your First Agent

This tutorial walks you through creating, running, and communicating with your first AI agent in mycel.

Prerequisites

  • The mycel server running (mycel up -d from your repo)
  • An AI provider configured (e.g., Claude Code or Gemini)

Step 1: Create an agent

Create an engineer agent named eng-01:

mycel agent create eng-01 --template engineer

This creates:

  • A git worktree under ~/.mycel/, checked out from the agent's repo
  • A Docker container (or tmux session) for the agent
  • Role-specific configuration files (CLAUDE.md, .mcp.json)

Step 2: Verify the agent is running

mycel status

You should see eng-01 listed with state idle:

AGENT     ROLE      STATE    UPTIME    TASK
eng-01    engineer  idle     10s       -

Step 3: Send work to the agent

mycel agent send eng-01 "Add a health check endpoint that returns JSON with status and uptime"

The agent receives the message in its session and begins working.

Step 4: Monitor progress

Watch the agent's recent output:

mycel agent peek eng-01

Or attach directly to the agent's tmux session:

mycel agent attach eng-01

Press Ctrl+B then D to detach from the session without stopping the agent.

Step 5: Subscribe to notifications

Subscribe the agent to a notification channel so it receives platform events:

# Show gateway connection status
mycel notify status

# List all agent subscriptions
mycel notify list

# Subscribe eng-01 to a Slack channel
mycel notify subscribe slack:engineering eng-01

# Show recent delivery activity
mycel notify activity slack:engineering --limit 10

Step 6: Check costs

After the agent has been working for a while, check spending:

mycel cost show
mycel cost agent

Step 7: Stop the agent

When the work is complete:

mycel agent stop eng-01

To clean up entirely (removes worktree and state):

mycel agent delete eng-01

Next steps