You've got Claude Code set up. Good. But right now it's like a really smart contractor you have to call every time you want something done. It doesn't remember yesterday's conversation. It doesn't run on its own. It waits for you.
OpenClaw changes that. It turns Claude Code into a full-time employee. An agent that remembers everything, runs on schedules, accesses your tools & handles complex workflows without you being there.
This is how Jars works. My AI agent that does morning briefings, CRM management, lead outreach, overnight code builds & research for DMpro. All built on OpenClaw.
What is OpenClaw?
OpenClaw is an open-source framework for building persistent AI agents. Think of it as the operating system your AI employee runs on.
Without OpenClaw, Claude Code is a one-off tool. With it, you get:
- Persistent memory: Your agent remembers past conversations, decisions & context
- Tool integrations: Connect Gmail, Slack, CRM, calendars, databases & more
- Scheduled runs: Set your agent to run at 7am, every hour, whenever you want
- Multi-step workflows: Complex tasks that require multiple actions & decisions
- Multi-channel delivery: Get results via Slack, email, SMS, webhooks
(instructions)
(orchestrator)
(brain)
(Gmail, CRM...)
Prerequisites
Before starting, make sure you have:
- Claude Code installed & working
- Git installed (comes with most computers, or download from git-scm.com)
- An Anthropic API key (from console.anthropic.com)
If you don't have Claude Code yet, follow the setup guide first. Takes 15 minutes.
Step 1: Clone OpenClaw
Open your terminal and run:
You've got the source code. Now let's configure it.
Step 2: Configure your environment
Copy the example environment file & add your keys:
Open the .env file in any text editor and add:
# Your Anthropic API key
ANTHROPIC_API_KEY=sk-ant-your-key-here
# Agent name (whatever you want)
AGENT_NAME=my-agent
# Where to store agent memory
MEMORY_PATH=./memory
# Notification channel (optional)
SLACK_WEBHOOK=https://hooks.slack.com/your-webhook
The only required field is your Anthropic API key. Everything else has sensible defaults.
Step 3: Create your first agent config
This is where it gets good. An agent config is basically a job description written in plain English. Create a file called agents/morning-briefing.yaml:
# agents/morning-briefing.yaml
name: "morning-briefing"
schedule: "0 7 * * *" # Every day at 7:00 AM
persona: |
You are my morning briefing agent.
You're efficient, direct & don't waste words.
Think of yourself as my chief of staff.
instructions: |
Every morning, do the following:
1. Check Gmail for unread emails. Flag anything urgent.
2. Check my CRM for new leads & deal updates.
3. Review today's calendar.
4. Check yesterday's metrics (revenue, signups).
5. Compile everything into a clean briefing.
6. Send to Slack #morning-briefing channel.
Format: bullet points, grouped by category.
Keep it under 300 words. I read this on my phone.
tools:
- gmail
- google-calendar
- hubspot
- slack
- stripe
memory: true
max_tokens: 4000
That's your agent. Notice there's zero code in there. It's all natural language describing what you want the agent to do, what tools it can use & when to run.
Step 4: Set up tool integrations
Your agent needs credentials to access your tools. OpenClaw makes this straightforward:
Each tool walks you through its setup. Most use OAuth (click to authorize) or API keys (copy-paste). Takes about 5 minutes per tool.
Step 5: Test your agent
Before going live, run a test:
The --dry-run flag means it does everything except send the message. Check the output. Is it useful? Tweak the instructions until it's perfect.
Step 6: Go live
Happy with the test? Remove the dry-run flag and activate the schedule:
Done. Tomorrow morning at 7am, your agent wakes up, scans everything & delivers your briefing. Every single day. Without you lifting a finger.
Understanding agent memory
This is the part that makes OpenClaw special. Your agent remembers things.
After a few days, it learns patterns. It knows which emails you actually care about. It tracks which leads are progressing. It notices when metrics are unusual.
Memory is stored in markdown files in the memory/ folder. Your agent reads them every session and writes new observations after each run. You can even edit them manually to give your agent extra context.
# memory/2025-01-15.md
## Morning Briefing Notes
- Client Acme Corp mentioned Q2 contract renewal
- Lead from Twitter DM campaign converting well
- Revenue trending 12% above last week
- Johann prefers briefings under 200 words
This is basically how my agent Jars became so good. Over months of memory, it understands my business better than most humans would.
Building more agents
Once you've got the morning briefing running, the next agents are easy. Same pattern, different job:
🎯 Lead Outreach Agent
Scans CRM for hot leads, writes personalized outreach messages, queues them for optimal send times. This is how Jars generated $48k in pipeline overnight.
🔄 Follow-up Agent
Monitors deals in your pipeline. If a deal goes quiet for 3 days, it drafts a follow-up. If someone opened your proposal, it flags them as hot.
🌙 Overnight Build Agent
Before bed, describe what you want built. The agent works through the night, shipping code & deploying changes. Wake up to a finished feature.
🔍 Research Agent
Monitors competitors, tracks industry news, finds keyword opportunities & summarizes findings. My agent does this for SEO research on DMpro.
Tips for better agents
Write detailed personas
Don't just say "you're my agent." Give it a personality. "You're my chief of staff. You're direct, efficient & you flag problems early. You don't sugarcoat." The more specific the persona, the better the output.
Be explicit about format
"Bullet points, grouped by category, under 300 words." This makes a huge difference. Without format instructions, agents tend to be verbose.
Use memory strategically
Edit memory files to give your agent context it can't discover on its own. "Our biggest client is Acme Corp. They renew in Q2. Always prioritize their emails."
Start simple, iterate
Your first agent config will be basic. That's fine. Run it for a week. See what's missing. Add more instructions. This is how you build something great.
Join the community
Inside AI Operators, you get:
- My exact OpenClaw agent configs (the ones running Jars)
- Pre-built templates for common agent types
- Tool integration guides for every popular service
- Direct help when you get stuck
- Other builders sharing their setups & results
The community is free. The templates alone save you hours of configuration. Learn more about the full process in our guide on how to build AI agents or see how it applies to automating your business.