Your First OpenClaw Agent in 10 Minutes

You don’t need to understand AI deeply to build a useful agent. You need three things: a personality file, a workspace, and a schedule. That’s it. Let’s do it in 10 minutes.

What You’ll Build

A personal assistant agent named “Aria” that:

Lives in its own isolated workspace (~/agents/aria/)
Has a distinct personality and purpose (defined in SOUL.md)
Runs a daily briefing every morning at 8 AM
Can be triggered manually via Telegram
Remembers context across sessions using memory files

npm install -g openclaw
openclaw --version  # Verify: should print 1.x.x or higher

# Run inside WSL2 Ubuntu
npm install -g openclaw
openclaw --version

Then configure your API keys:

openclaw init
# Follow the prompts:
# - Anthropic API key (required)
# - Telegram bot token (optional but recommended)
# - Default model (press Enter for claude-opus-4 default)

This creates ~/.openclaw/config.json — the global config all agents share.

Step 2: Create the Agent Workspace

Every agent lives in its own directory. This isolation keeps agents from stepping on each other and makes them easy to back up or share.

mkdir -p ~/agents/aria
cd ~/agents/aria

# Create the memory directory structure
mkdir -p memory/entities/{people,companies,projects}
mkdir -p state

Your workspace structure will look like:

~/agents/aria/
├── SOUL.md          ← Personality & purpose (you'll write this)
├── MEMORY.md        ← Long-term curated memory (starts empty)
├── AGENTS.md        ← Workspace rules (optional but useful)
├── memory/
│   ├── 2026-03-27.md    ← Daily logs (auto-created)
│   └── entities/        ← Structured facts about people/companies
└── state/
    └── current.md       ← Active context (survives compaction)

Step 3: Write Your Agent’s Personality

SOUL.md is the single most important file. It defines who your agent is — its values, communication style, and what it cares about. Agents without a SOUL.md are generic. Agents with one are memorable.

Create ~/agents/aria/SOUL.md:

# SOUL.md - Aria's Personality

## Who I Am

I'm Aria, a personal research and briefing assistant.
My job is to keep you informed without drowning you in noise.

## Core Values

**Be concise.** No padding, no filler. If something can be said in
10 words instead of 30, use 10.

**Be curious.** When I find something interesting, I say so. I have
opinions about ideas, not just neutral summaries.

**Be proactive.** I flag things before you ask. If I see something
relevant, I surface it — even if you didn't request a report.

## Communication Style

- Short paragraphs, no walls of text
- Use bullets for lists, prose for narrative
- Never say "Great question!" or "I'd be happy to help!"
- Include a bottom-line-up-front (BLUF) when summarizing

## What I Focus On

- AI and technology news
- Business and strategy insights
- Anything directly relevant to your current projects

## What I Ignore

- Celebrity news, sports, entertainment
- Clickbait or sensationalism
- Anything I've already reported in the last 7 days

Step 4: Register the Agent in openclaw.json

OpenClaw needs to know Aria exists. Edit your main OpenClaw config at ~/.openclaw/openclaw.json (or ~/clawd/openclaw.json if you’re using the standard setup):

{
  "agents": {
    "main": {
      "workspace": "~/clawd",
      "model": "anthropic/claude-opus-4",
      "description": "Main personal assistant"
    },
    "aria": {
      "workspace": "~/agents/aria",
      "model": "anthropic/claude-haiku-4",
      "description": "Daily research briefing agent",
      "channel": "telegram"
    }
  }
}

Key fields explained:

workspace: Where the agent’s files live (absolute or ~/ paths work)
model: Which LLM to use. Haiku is 10× cheaper than Opus for scheduled tasks
description: Human-readable label — shows up in dashboards and logs
channel: Where the agent sends output by default

Step 5: Create Your First Cron Schedule

Cron schedules let agents run automatically. Add Aria’s morning briefing to your cron config:

openclaw.json (recommended)
Cron syntax reference

{
  "agents": {
    "aria": {
      "workspace": "~/agents/aria",
      "model": "anthropic/claude-haiku-4",
      "description": "Daily research briefing agent",
      "channel": "telegram",
      "crons": [
        {
          "schedule": "0 8 * * *",
          "prompt": "Read SOUL.md and memory/MEMORY.md. Then search the web for the top 3 AI and technology stories from the last 24 hours. Write a concise morning briefing (3-5 bullets, 1-2 sentences each) and send it to the telegram channel. Log what you sent in memory/YYYY-MM-DD.md.",
          "model": "anthropic/claude-haiku-4",
          "timeout": "5m"
        }
      ]
    }
  }
}

┌──────── minute (0-59)
│ ┌────── hour (0-23)
│ │ ┌──── day of month (1-31)
│ │ │ ┌── month (1-12)
│ │ │ │ ┌ day of week (0-6, 0=Sunday)
│ │ │ │ │
0 8 * * *    → Every day at 8:00 AM
0 9 * * 1    → Every Monday at 9:00 AM
*/30 * * * * → Every 30 minutes
0 8,17 * * * → 8 AM and 5 PM daily

Step 6: Test via Telegram

Before trusting the cron, test manually:

# Trigger Aria directly
openclaw run aria --prompt "Quick test: search the web for 1 AI news story from today and send it to me on Telegram"

# Or open an interactive session
openclaw chat aria

You should see Aria respond in Telegram within 30-60 seconds. If you don’t:

# Check logs
openclaw logs aria --tail 50

# Verify config
openclaw status aria

Common issues:

No Telegram message: Check bot token and that you’ve messaged the bot first (Telegram requires users to initiate)
Model errors: Verify your Anthropic API key has credits
“Agent not found”: Double-check the agent name matches exactly in openclaw.json

Step 7: Verify the Cron

# List scheduled jobs
openclaw crons list

# Output:
# aria     0 8 * * *   Daily briefing    Next: 2026-03-28 08:00:00

To test the cron fires correctly without waiting until 8 AM:

openclaw crons trigger aria
# Fires the cron immediately with its configured prompt

What You Built

Isolated workspace

Aria lives in ~/agents/aria/ — her own files, memory, and state. She can’t accidentally read or modify other agents’ data.

Defined personality

SOUL.md gives Aria a consistent voice and scope. Every session, she reads it first and stays in character.

Memory across sessions

Daily memory files mean Aria remembers what she reported yesterday. No duplicate briefings.

Automated schedule

The 8 AM cron runs without you touching anything. Aria works whether your laptop is open or not (if running on a server).

Next Steps

Now that Aria is running, extend her:

# Add to the cron prompt to prevent repeats:
"Before searching, read memory/YYYY-MM-DD.md from the last 3 days.
Skip any stories you've already reported."

# Add a second cron for evening summary:
{
  "schedule": "0 17 * * 1-5",
  "prompt": "Check my calendar for tomorrow's events and send a 3-bullet prep note."
}

Ready to build something more complex? Check out Multi-Agent Orchestration: CEO Pattern to learn how agents can delegate to each other.

About the author: JD Davenport builds AI agent systems at OpenClaw. Follow on LinkedIn for updates on building AI agents for business.