Step 1 — Install

Download & Install OpenClaw

OpenClaw runs on macOS, Windows, and Linux. Pick your platform and run the installer — no Docker required, no dependencies to chase down.

Terminal

            # Option A — Homebrew (recommended)
brew install openclaw

# Option B — Direct download
curl -fsSL https://get.openclaw.dev | sh

# Verify install
openclaw --version

PowerShell (run as Administrator)

# Option A — winget
winget install OpenClaw.OpenClaw

# Option B — manual installer
# Download OpenClaw-Setup.exe from openclaw.dev/download
# Run the installer and follow the wizard

# Verify install (new terminal)
openclaw --version

Terminal

            # Debian / Ubuntu
curl -fsSL https://get.openclaw.dev | sh

# Arch Linux
yay -S openclaw

# Verify install
openclaw --version

First launch OpenClaw creates a config folder at ~/.openclaw/ on macOS/Linux or %APPDATA%\OpenClaw\ on Windows. This is where your SOUL.md, memory database, and skill files live.

Step 2 — Model

Connect Your AI Model

OpenClaw is model-agnostic. You can use a cloud API or run a local model entirely on your own hardware. Pick what fits your budget and privacy needs.

Claude (Anthropic)

Best reasoning and long-context handling. Recommended for most users.

API KeySmartest

GPT-4o (OpenAI)

Strong all-rounder with excellent tool-use and vision capabilities.

API KeyFast

Gemini (Google)

Massive context window. Great for processing very long documents.

API KeyFast

Ollama (Local)

Run Llama 3, Mistral, Phi-3 and others locally. 100% private, no API costs.

FreeLocal

~/.openclaw/config.yaml

            
            model:
  provider: "anthropic"        # anthropic | openai | google | ollama
  model_id: "claude-sonnet-4-6"
  api_key:  "sk-ant-..."          # paste your key here

memory:
  backend: "sqlite"              # sqlite (default) | postgres

gateway:
  whatsapp: false               # enable later in Step 6
  telegram: false
            

Keep your API key safe. Never commit config.yaml to a public Git repo. OpenClaw also supports reading the key from an environment variable: OPENCLAW_API_KEY=sk-ant-... openclaw start

Step 3 — Identity

Write Your SOUL.md

SOUL.md is a plain Markdown file that tells OpenClaw who it is, how it should behave, and what it should prioritise. It's loaded into the system prompt on every run — think of it as your agent's personality and standing orders.

# ~/.openclaw/SOUL.md ## Identity You are Alex, a focused personal assistant for a freelance designer. Your job is to save time, reduce busywork, and surface what matters. ## Tone - Concise and direct. No fluff, no filler sentences. - Friendly but professional. Use first names when appropriate. - If you're unsure about something, say so — don't guess silently. ## Priorities 1. Client work always comes first. 2. Respond to any message tagged [URGENT] before anything else. 3. Never delete files without explicit confirmation. ## Default behaviours - Summarise emails into 3 bullets unless I ask for full text. - Always save research findings to memory before reporting back. - Format code in markdown code blocks with language tags. ## Off-limits - Do not post to social media without my explicit approval. - Do not share files outside my local machine without asking.

Start simple. Even a five-line SOUL.md dramatically improves output quality over a blank one. You can always expand it later — OpenClaw reloads SOUL.md on every Heartbeat cycle, so changes take effect within seconds.

Step 4 — First Task

Run Your First Task

Start OpenClaw, then give it a task either through the terminal, the web UI at localhost:7799, or WhatsApp/Telegram if you've connected them. Here's a classic first task to see every part of the system in action:

Your message to OpenClaw

"Search the web for the 3 most important AI news stories from this week. Summarise each in 2 sentences and save them to memory. Then send me a formatted report."

What OpenClaw does

Calls web_search tool — searches for recent AI news

Calls read_page tool — reads the top 3 articles

Calls memory_write — saves summaries to SQLite

Formats a Markdown report and returns it

Heartbeat logs the run — auditable in ~/.openclaw/logs/

Result

A formatted three-item report appears in your terminal (or message app) in about 20–40 seconds. The summaries are also stored in memory and available for future tasks.

Starting OpenClaw

            
            # Start the agent (runs in background)
openclaw start

# Send a one-off task from terminal
openclaw run "summarise my 5 most recent emails"

# Open the web UI
openclaw ui
# → opens http://localhost:7799 in your browser

# Watch live logs
openclaw logs --follow
            

It worked if… you see a response in the terminal or web UI, and openclaw memory list shows at least one entry. Congratulations — your agent is alive.

Step 5 — Skills

Install Your First Skill

Skills are pre-packaged capabilities you install from ClawHub. They give OpenClaw new tools it doesn't have out of the box — like sending Slack messages, reading Google Sheets, or generating images.

Browse ClawHub

Run openclaw skills browse or visit the Marketplace page to find skills. For your first one, try gmail-reader or daily-briefing.

Install the skill

Run openclaw skills install daily-briefing. OpenClaw downloads the skill file and verifies its signature against ClawHub's registry.

Configure it (if needed)

Some skills need credentials. Run openclaw skills configure daily-briefing to step through the setup wizard — it stores secrets securely in the system keychain, never in plain text.

Use it in a task

Skills become available immediately. Just reference them naturally: "Use the daily-briefing skill to send me my morning summary at 8 AM."

Common skill commands

            
            openclaw skills browse         # list all available skills
openclaw skills install <name>  # install a skill
openclaw skills list            # show installed skills
openclaw skills update <name>   # update a skill
openclaw skills remove <name>   # uninstall a skill
openclaw skills info <name>     # show skill details and permissions
            

Step 6 — Automation

Set Up Your First Automation

OpenClaw can run tasks on a schedule, in response to triggers, or when conditions are met — without you lifting a finger. Here are three ways to automate:

Scheduled task (cron-style)

            
            # Run a task every weekday at 8 AM
openclaw schedule add \
  --cron "0 8 * * 1-5" \
  --task "Summarise my overnight emails and Slack messages. Send me a briefing."
            

Trigger-based automation

            
            # Fire when a new file appears in ~/Downloads
openclaw trigger add \
  --event "file:created" \
  --path "~/Downloads/*.pdf" \
  --task "Summarise this PDF and save the summary to memory."
            

Gateway (WhatsApp / Telegram)

            
            # Enable messaging in config.yaml
gateway:
  telegram: true
  telegram_token: "your-bot-token"

# Now message your bot on Telegram and OpenClaw responds
            

Start with one scheduled task. A Monday morning briefing is the single highest-value automation for most new users — it compounds value every week with zero ongoing effort.

Common Setup Issues

Hit a snag? Here are the most frequent first-time problems and how to fix them.


Error	Likely cause	Fix
openclaw: command not found	PATH not updated after install	Restart your terminal, or add `~/.openclaw/bin` to your PATH manually
Authentication error (401)	Invalid or expired API key	Check `config.yaml` — ensure the key has no leading/trailing spaces and has billing enabled on your provider account
No response after 60 seconds	Model timeout or no internet	Run `openclaw diagnose` to check connectivity. Try a simpler task first to isolate the issue
SOUL.md not loaded	Wrong file location or encoding	File must be at `~/.openclaw/SOUL.md` and saved as UTF-8. Run `openclaw soul check` to validate
Skill install failed	Signature verification failed	The skill may have been tampered with. Only install skills from verified publishers on ClawHub. See the Security guide
Memory database locked	Two OpenClaw processes running	Run `openclaw stop` then `openclaw start` to restart cleanly

Still stuck? Check the FAQ or open an issue on the OpenClaw GitHub repo.

You're Up and Running — What's Next?

Now that OpenClaw is live, explore what it can really do.

Automation Guide Browse ClawHub Multi-Agent FAQ 📋 Command Cheat Sheet

Setup Steps