Your First 72 Hours With OpenClaw: The Setup Guide Everyone Skips

Hour 0-1: How Do You Get OpenClaw Running?

This is the foundation. Skip nothing here, because every subsequent step depends on a clean installation.

Minute 0-5: Choose your hosting. If you already have a VPS or home server, use it. If not, spin up a VPS now. Hetzner CAX11 (4 EUR/month) or a Hostinger KVM plan work well. You need at least 2GB RAM and 2 vCPUs. If you want to test locally first, your own computer works fine — just know that you will eventually want to move to a VPS for 24/7 uptime.

Minute 5-15: Install Docker. SSH into your server (or open a terminal locally) and install Docker and Docker Compose. On Ubuntu:

curl -fsSL https://get.docker.com | sh
sudo usermod -aG docker $USER

Log out and back in for the group membership to take effect.

Minute 15-25: Deploy OpenClaw. Clone the OpenClaw repository or create a docker-compose.yml with the official OpenClaw image. Set your environment variables: the AI model provider (Anthropic, OpenAI, or DeepSeek), your API key, and — critically — your gateway token. The gateway token is your first line of security. Set it now, not later.

docker compose up -d

Minute 25-35: Verify it works. Access the OpenClaw web UI (if enabled) or send a test message via the API. Confirm you get a response from the AI model. If you see a response, your core installation is working. If not, check the logs with docker compose logs — the most common issues are incorrect API keys, network configuration, or insufficient RAM.

Minute 35-60: Harden security. Before you do anything else, lock down your installation. Set the gateway token if you have not already. Configure your firewall to allow only necessary ports. Set up HTTPS with Caddy or Nginx. Review the 5 common setup mistakes and make sure you are not making any of them.

At the end of Hour 1, you should have: a running OpenClaw instance, a configured AI model, a gateway token set, HTTPS enabled, and confirmation that the agent responds to messages. This is your foundation.

Hour 1-4: How Do You Connect Your First Channel?

A channel is how people (including you) communicate with your agent. WhatsApp and Telegram are the two most popular first channels because they are on your phone, making the agent instantly accessible.

Choosing your first channel:

• WhatsApp (via WhatsApp Business API or Evolution API): Best if WhatsApp is your primary messaging app. Requires a phone number and some API configuration. More complex to set up but more useful for most people.
• Telegram: Easiest to set up. Create a bot via @BotFather, get a token, add it to your OpenClaw config. Working in under 10 minutes. Great for testing.
• Discord: Good if you already run a Discord server. Requires creating a bot application and inviting it to your server.

For your first channel, choose Telegram unless you have a strong reason to start elsewhere. It is the fastest to set up and lets you start testing immediately. You can add WhatsApp or Discord later.

Setup steps (Telegram):

1. Message @BotFather on Telegram and create a new bot. Save the token.
2. Add the Telegram bot token to your OpenClaw environment variables.
3. Restart OpenClaw to pick up the new configuration.
4. Send a message to your bot on Telegram.
5. Verify you receive a response.

Your first real conversation: Once the channel is connected, have a real conversation. Not just "hello" — ask your agent to do something useful. Ask it to summarize an article. Ask it a question about your industry. Ask it to draft an email. Get a feel for the response quality, speed, and tone. This is the baseline you will improve over the next two days.

At the end of Hour 4, you should have: at least one working messaging channel, confirmation that messages flow reliably in both directions, and a sense of the agent's baseline capabilities.

Day 1: How Do You Set Up Memory and Your First Skill?

Day 1 is when your agent starts becoming more than a chatbot. Memory and skills are the two features that unlock real value.

Setting up memory (1-2 hours):

Memory is how your agent retains information between conversations. Without it, every conversation starts from zero. Create a memory.md file in your OpenClaw data directory with the following sections:

• About you: Your name, role, business, key clients or projects, communication preferences.
• Key contacts: People the agent should know about — their names, roles, relationship to you, any special handling instructions.
• Business context: Your industry, products/services, pricing, common questions you receive.
• Agent instructions: Tone of voice, formatting preferences, things the agent should always or never do.

Start with 500-1000 words of memory. You will add to it over time. The goal right now is to give the agent enough context to be useful in your specific situation, not to document everything about your life.

Installing your first skill (30-60 minutes):

Skills are task-specific instructions that tell your agent how to handle specific types of requests. Your first skill should solve a real, recurring problem for you. Good first skills include:

• Email drafting: Takes a brief and produces a draft in your tone.
• Meeting summary: Takes notes and produces a structured summary with action items.
• Content repurposing: Takes long-form content and adapts it for a specific platform.

Browse ClawHub for a community skill that matches your need, or write a simple SKILL.md file. A skill file is just a markdown document that describes the task, the expected input, the desired output format, and any constraints. Start simple. You can make it more sophisticated later.

At the end of Day 1, you should have: a memory file with your key information, and at least one working skill that the agent uses correctly when triggered.

Day 2: How Do You Add Cron Jobs and a Morning Briefing?

Day 2 is when your agent becomes proactive. Until now, it only responds when you talk to it. Cron jobs let it take action on a schedule — and this is where the real magic happens.

Your first cron job: the morning briefing (1-2 hours):

A morning briefing is the single most valuable cron job you can set up. It runs at a set time (say, 7:00 AM), gathers information from your connected services, and sends you a summary message on your preferred channel.

Start simple. Your first morning briefing might just include:

• Today's calendar events
• Any tasks or reminders you set
• A brief weather summary
• Any pending follow-ups from yesterday

To set it up, create a cron skill that defines what information to gather and how to format it. Then configure a cron schedule in OpenClaw to trigger this skill at your desired time. The exact configuration depends on your OpenClaw version, but the concept is the same: trigger, skill, channel.

After the morning briefing runs for the first time, review the output. Is the format useful? Is anything missing? Is anything included that you do not need? Iterate on the skill definition until the briefing feels like it was written by an assistant who knows you.

Optional: Connect a second channel (30-60 minutes):

If you started with Telegram for testing, now is a good time to add WhatsApp or Discord. Having multiple channels lets you use different platforms for different contexts — Telegram for quick personal queries, WhatsApp for client-facing communication, Discord for team collaboration.

At the end of Day 2, you should have: at least one cron job running on schedule (ideally a morning briefing), and optionally a second messaging channel configured.

Day 3: How Do You Review, Tune, and Add Monitoring?

Day 3 is about quality, not features. Resist the urge to add more skills and integrations. Instead, focus on making what you have work well.

Review conversation quality (30-60 minutes):

Read through all the conversations your agent has had over the past two days. Look for:

• Incorrect information: Did the agent state anything wrong? If so, update your memory files or skill definitions to prevent it.
• Tone issues: Does the agent sound like you, or does it sound generic? Adjust your agent instructions in memory.
• Missed context: Did the agent fail to use information that was in its memory? Check that your memory files are structured clearly and that memory search is working.
• Skill failures: Did any skills produce poor output? Review the skill definition and add more specific instructions.

Tune your morning briefing (30 minutes):

You have had one or two morning briefings by now. What worked? What was missing? What was included but not useful? Revise the briefing skill based on real-world experience. Most people go through 3-5 iterations before the briefing format feels right.

Set up basic monitoring (30-60 minutes):

• Sign up for UptimeRobot (free tier) and add a monitor for your OpenClaw instance URL.
• Check your API provider's usage dashboard and note your current spending rate.
• Set a daily spending alert if your provider supports it.
• Enable OpenClaw's built-in logging for API calls.

Document what is working (15 minutes):

Before the details fade, write a quick note about what is working, what is not, and what you want to improve next week. This becomes your roadmap for Week 2.

At the end of Day 3, you should have: a reviewed and tuned agent, a morning briefing that delivers real value, basic monitoring in place, and a clear sense of what to improve next.

What Should You NOT Do in the First 72 Hours?

The 460-point Reddit post that inspired this guide had one consistent theme in the comments: people who tried to do too much too fast burned out and quit. Here is what to avoid:

Do not connect more than two channels. Each channel adds complexity. Master one or two before adding more.

Do not install more than three skills. Each skill adds to the context window and increases the chance of skill conflicts. Start with one, add a second when the first is working well, and add a third only if you genuinely need it.

Do not give write or delete permissions on any integration. For the first 72 hours, your agent should be read-only on all connected services. Let it observe and report. You can add action permissions in Week 2 after you trust its judgment.

Do not skip security hardening. It takes 30 minutes. Not doing it can cost you thousands in compromised API credits. There is no scenario where skipping security is worth it.

Do not expect perfection. Your agent will give wrong answers. It will format things oddly. It will miss context. This is normal. The first 72 hours are about building a foundation that you improve over time, not about achieving a perfect setup on day one.

Do not compare your Day 1 agent to someone's Month 6 agent. The people posting impressive OpenClaw setups on Reddit have been iterating for months. Their agents have rich memory, tuned skills, and dozens of cron jobs. Yours will get there too — just not in 72 hours.

The first 72 hours are the most critical period. If you follow this guide, you will have a solid foundation that you can build on for months. If you try to skip ahead, you will end up with a fragile mess that breaks in unpredictable ways. Take it step by step, and by Day 3 you will understand why people get excited about OpenClaw.