How to Set Up OpenClaw in 2026: Complete Beginner’s Guide

ByL.B. Kaelin

This is the definitive beginner’s guide for how to set up OpenClaw in 2026. By the end, you will have a running OpenClaw agent on your own machine or VPS, connected to Slack or Telegram, with persistent memory that survives restarts, and a scheduled daily briefing cron job. We wrote this for real beginners — people who have heard about AI agents but have not set one up before. You do not need to be a developer, but you should be comfortable running commands in a terminal. If you can install software on your computer, you can set up OpenClaw.

What OpenClaw Is (And Why It’s Different)

OpenClaw is an open-source AI agent platform that lets you run persistent, autonomous agents on your own machine or a VPS. Where ChatGPT and Claude are chat interfaces that forget everything when you close the tab, OpenClaw agents remember every conversation, run on schedules, and connect directly to Slack, Telegram, and Discord.

Here is the simplest way to understand the difference:

  • ChatGPT — zero memory between sessions unless you manually copy-paste. No automation. No integration with your tools. It is a chat window.
  • Claude Desktop — better context window, but still no scheduling, no cron jobs, no persistent agent identity. You talk to it, it responds, the conversation ends.
  • OpenClaw — the agent has a permanent identity defined in a file called SOUL.md. It remembers everything in MEMORY.md. It can post to Slack at 8 AM every day automatically. It can run tasks on a schedule while you sleep. It lives on your machine, not in someone’s cloud.

OpenClaw is also the fastest-growing open-source AI project in the world as of April 2026 with over 347,000 GitHub stars and an estimated 3 million active users. It runs on Mac, Linux, and Windows (via WSL2). You bring your own LLM API key, so you control costs and model choice.

What You Need Before You Start

System Requirements

  • Operating System: macOS (Apple Silicon M1+ or Intel), Linux (Ubuntu 22.04 or 24.04 recommended), or Windows with WSL2
  • RAM: 2 GB minimum, 4 GB recommended
  • Storage: 500 MB for OpenClaw itself, plus space for your agent’s memory and workspace files
  • Node.js: Version 18 or higher required. Version 22 is strongly recommended. You can check your current version by running node -v in a terminal.

API Key

OpenClaw does not include built-in LLM access. You bring your own API key from one of these providers:

  • OpenAI — GPT-4o, GPT-4o mini. Best quality but highest cost. Start with GPT-4o mini for testing.
  • Anthropic — Claude Sonnet, Claude Haiku. Excellent for long-form reasoning.
  • DeepSeek — DeepSeek V3. Excellent quality, very low cost. Our recommended model for beginners.
  • OpenRouter — One API key gives you access to dozens of models. Pay as you go.
  • Ollama — Run models locally on your own hardware. Free (no API costs) but requires a machine with sufficient GPU or RAM.

For beginners, we recommend starting with DeepSeek V3 or GPT-4o mini. Both are cheap, fast, and good enough for testing while you learn. You can switch to a more powerful model later.

Step 1: Install OpenClaw

Open a terminal and run this command:

npm install -g openclaw

This installs OpenClaw globally on your system. The installation takes about 30 seconds on a typical internet connection. If you see permission errors on Linux or macOS, you may need to prefix with sudo:

sudo npm install -g openclaw

Verify the installation worked:

openclaw --version

You should see a version number printed. If you see “command not found,” your Node.js global bin directory may not be in your PATH. Restart your terminal and try again.

Docker Alternative

If you prefer Docker or do not want to install Node.js directly, run:

docker run -p 3000:3000 openclaw/openclaw

This pulls the official OpenClaw Docker image and starts it on port 3000. All subsequent steps work the same way regardless of installation method.

Step 2: Run the Setup Wizard

Start the OpenClaw daemon:

openclaw start

This launches the OpenClaw server in the background. Your terminal will show log output. Wait until you see a message that the server is running, then open your browser and go to:

http://localhost:3000

The OpenClaw web UI will open. On first launch, it walks you through a setup wizard:

  1. Select your LLM provider — Choose from the list (OpenAI, Anthropic, DeepSeek, OpenRouter, Ollama, or others).
  2. Enter your API key — Paste your key from the provider. The key stays on your machine. OpenClaw never sends it anywhere except to the LLM provider you selected.
  3. Choose a workspace directory — This is where your agent’s files live. The default is ~/.openclaw/workspace. You can change this if you want.
  4. Name your agent — Give it a name. This is optional but recommended.

When the wizard finishes, it creates ~/.openclaw/openclaw.json. This is the main configuration file for your OpenClaw instance. Every setting — from which model to use, to which Slack channel to connect, to what cron jobs to run — lives in this file.

You can open openclaw.json in any text editor to see what the wizard generated. It looks something like this (values will differ based on your choices):

{
  "llm": {
    "provider": "deepseek",
    "model": "deepseek-chat",
    "apiKey": "sk-..."
  },
  "workspace": {
    "dir": "/home/you/.openclaw/workspace"
  }
}

Step 3: Set Up Your Workspace Files

OpenClaw uses three files in your workspace directory to define how your agent behaves and what it remembers. Understanding these files is the key to getting value from OpenClaw.

SOUL.md — Your Agent’s Identity

SOUL.md defines who your agent is. Without it, your agent has no personality or role. It will give generic, unhelpful responses. With a good SOUL.md, your agent sounds like a real assistant.

Here is a simple SOUL.md for a personal productivity assistant:

# Personal Assistant

## Identity
You are Alex, a helpful personal assistant. You are organized, friendly, and direct. You never waste time with fluff.

## Role
You help with task management, note-taking, research, and daily planning. You keep track of deadlines and priorities.

## Communication
Be concise. Use bullet points for lists. If something is urgent, say so clearly. Ask clarifying questions when the user's request is ambiguous.

Create this file at ~/.openclaw/workspace/SOUL.md and customize it to your needs.

MEMORY.md — Persistent Long-Term Memory

MEMORY.md is where your agent stores everything it learns. It is append-only — the agent adds new information to the end of the file but never overwrites old entries. This is what makes OpenClaw different from ChatGPT: your agent remembers who you are, what you talked about last week, and what projects you are working on.

OpenClaw can read and write to MEMORY.md automatically if you tell it to in AGENTS.md. You do not need to create MEMORY.md manually — the agent will create it on first use — but you should verify the file exists after your first conversation.

AGENTS.md — Operational Protocols

AGENTS.md tells your agent how to operate. This is where you instruct it to check MEMORY.md at the start of every session, follow specific workflows, or run in a particular style.

Here is a minimal AGENTS.md that ensures your agent uses persistent memory:

# Operational Instructions

## Session Start
1. Read SOUL.md
2. Read MEMORY.md
3. Greet the user briefly
4. Ask if there are priority items from memory

## Session End
After every conversation, write a brief summary to MEMORY.md
including the date and key topics discussed.

These three files are the foundation of every good OpenClaw setup. Take 10 minutes to write them well and your agent will feel like a completely different tool.

Step 4: Connect to Slack or Telegram (Optional but Recommended)

Connecting OpenClaw to a messaging platform means you can talk to your agent without opening the web UI. You can message it from your phone, and it can send you updates automatically.

Slack Setup

  1. Go to https://api.slack.com/apps and create a new app.
  2. Choose “From scratch,” give it a name, and select your workspace.
  3. Under “Add features and functionality,” enable Bots.
  4. Under “OAuth & Permissions,” add these Bot Token Scopes: chat:write, channels:history, channels:read, im:history, im:read, im:write, users:read.
  5. Install the app to your workspace and copy the Bot User OAuth Token.
  6. Add this token to your openclaw.json under the plugins section:
{
  "plugins": {
    "entries": {
      "slack": {
        "type": "slack",
        "config": {
          "token": "xoxb-your-token-here"
        }
      }
    }
  }
}
  1. Restart OpenClaw with openclaw restart.
  2. Open Slack and message your bot. It will respond in the channel.

Telegram Setup

  1. Open Telegram and search for @BotFather.
  2. Send /newbot and follow the prompts to create a bot. BotFather will give you an API token.
  3. Add the token to your openclaw.json:
{
  "plugins": {
    "entries": {
      "telegram": {
        "type": "telegram",
        "config": {
          "token": "your-telegram-bot-token"
        }
      }
    }
  }
}
  1. Restart OpenClaw with openclaw restart.
  2. Find your bot on Telegram and send /start. You can now message it directly.

Discord setup follows a similar pattern through the Discord Developer Portal. The plugin system in openclaw.json handles all three platforms.

Step 5: Test Your Agent

Now that everything is configured, it is time to verify it works:

  1. Send a message in the web UI — Open http://localhost:3000 and type “Hello, what is my name and what did we talk about last time?” If you have MEMORY.md configured, the agent should check it and tell you if there is nothing yet.
  2. Send your bot a message on Slack or Telegram — Say “Remember that I like coffee with oat milk.” The agent should respond and write this to MEMORY.md.
  3. Verify memory persistence — Close your browser, wait 10 seconds, and open the web UI again. Ask “What do I like to drink?” The agent should recall “coffee with oat milk.” If it does not remember, check that AGENTS.md instructs it to read MEMORY.md at session start.
  4. Check your MEMORY.md file — Open ~/.openclaw/workspace/MEMORY.md and look for the log of your conversation. If it is empty or missing, your agent is not writing to it. Update AGENTS.md to include explicit instructions for memory management.

If all four checks pass, congratulations. You have a working OpenClaw agent with persistent memory connected to your messaging platform.

Step 6: Set Up Your First Cron Job (Optional)

One of the most powerful features of OpenClaw is the ability to run agents on a schedule. Cron jobs let your agent perform tasks automatically at specific times without any human triggering them.

Open your openclaw.json file and add a crons section. Here is a simple daily briefing that runs every weekday at 8:00 AM:

{
  "crons": [
    {
      "name": "daily-briefing",
      "schedule": "0 8 * * 1-5",
      "prompt": "Check MEMORY.md for any pending tasks from yesterday. Give me a brief morning update with: the current date, any deadlines approaching in the next 3 days, and one suggestion for what I should prioritize today.",
      "channel": "slack"
    }
  ]
}

The schedule field uses standard cron syntax. If you are unfamiliar with cron expressions, here are some common patterns:

  • 0 8 * * * — Every day at 8:00 AM
  • 0 9 * * 1 — Every Monday at 9:00 AM
  • */30 * * * * — Every 30 minutes
  • 0 0 1 * * — First day of every month at midnight

After adding cron entries, restart OpenClaw:

openclaw restart

Your agent will now run the task automatically at the scheduled time and deliver the result to the specified channel. You can have multiple cron jobs running different tasks — a morning briefing, an afternoon check-in, a weekly summary, or a nightly research session.

Common Setup Mistakes (And How to Fix Them)

Mistake 1: Empty or Missing SOUL.md

Symptom: Your agent gives generic, robotic responses. It does not sound like anyone in particular. It has no personality.

Fix: Write a SOUL.md file in your workspace directory. Even a simple one with a name and a role makes a dramatic difference. The 10 minutes you spend on this will change everything about how your agent feels.

Mistake 2: Using an Expensive Model for Testing

Symptom: Your first day of playing with OpenClaw costs $5 in API fees.

Fix: Start with DeepSeek V3 or GPT-4o mini. Both cost pennies per session. Switch to Claude Sonnet or GPT-4o only after you have everything configured and are doing real work. Your API bill will thank you.

Mistake 3: No Persistent Memory

Symptom: Every conversation starts fresh. Your agent does not remember anything you told it.

Fix: Set up AGENTS.md with explicit instructions to read MEMORY.md at session start and write to it at session end. Without these instructions, the agent does not know it should use memory. See the AGENTS.md example in Step 3 above.

Mistake 4: Wrong Node.js Version

Symptom: OpenClaw fails to start or throws cryptic JavaScript errors.

Fix: Check your Node.js version with node -v. If it is below v18, upgrade. If you are on v20 or v22, you are fine. Version 22 is the current stable release and is recommended for the best experience.

Mistake 5: Not Restarting After Config Changes

Symptom: You edited openclaw.json but nothing changed.

Fix: Run openclaw restart after every config change. OpenClaw reads openclaw.json at startup, not continuously.

Mistake 6: Forgetting to Invite the Bot to a Channel

Symptom: Your Slack bot is connected but does not respond in your channel.

Fix: In Slack, type /invite @your-bot-name in the channel where you want the bot to listen. Bots cannot join channels without being invited.

What to Do Next

Once you have the basics working, you can take your OpenClaw setup much further:

  • Upgrade your model — Switch from DeepSeek V3 or GPT-4o mini to Claude Sonnet or GPT-4o for more capable reasoning on complex tasks.
  • Add skills — Skills are pre-built capabilities your agent can use, like weather forecasting, web research, or code analysis. Browse the available skills and add them to your workspace.
  • Explore subagents — OpenClaw can spawn subagents for parallel tasks. A subagent can research a topic while your main agent handles another conversation. This is how power users run complex workflows.
  • Set up multiple agents — You can run different agents for different roles: a writing assistant, a coding assistant, a research agent. Each gets its own SOUL.md and MEMORY.md.
  • Deploy on a VPS — Once you trust your setup, deploy it on a $5/month VPS so it runs 24/7. Your agent will be online even when your laptop is closed.
  • Read the related guides on this site for deeper dives into specific topics.

Sources

  • OpenClaw Official Documentation — https://docs.openclaw.ai
  • OpenClaw GitHub Repository — https://github.com/openclaw/openclaw
  • Node.js Download Page — https://nodejs.org
  • Slack API Documentation — https://api.slack.com
  • Telegram Bot API — https://core.telegram.org/bots/api

Related Reading

Similar Posts