Give It Real Tasks

By the end of this page, your agent will be able to browse URLs, read and write files in your workspace, and execute shell commands — with your permission.

Time: ~20 minutes

What "real tasks" means

Up to now your agent has been a smart responder — it can answer questions, draft text, and reason. But a personal assistant should be able to do things:

• Fetch and summarize a URL you send
• Read a file and give you notes on it
• Save output to a file you can use later
• Run a command and report the result

These are called tools. OpenClaw's built-in tools are off by default. This page shows you how to turn them on.

Built-in tools

Add a tools section to ~/.openclaw/openclaw.json:

{
  "tools": {
    "web_fetch": { "enabled": true },
    "web_search": { "enabled": true },
    "read": { "enabled": true },
    "write": { "enabled": true },
    "exec": {
      "enabled": true,
      "allowlist": ["ls", "cat", "echo", "git status", "git log"]
    }
  }
}

What each tool does:

The exec tool is powerful. Use an allowlist to restrict which commands your agent can run. Never set allowlist: [] (empty = no restriction).

Restart and verify

openclaw gateway restart

Try these tasks

Once tools are enabled, test them in your messaging app:

Web fetch:

"Summarize this for me: https://example.com/article"

Your agent will fetch the URL and give you a summary.

Web search:

"What's the current price of AAPL?"

Your agent will search and return a result. (For real-time financial data with alerting, see the Finance Alerts goal.)

File write:

"Write a one-page project brief for [project] and save it as project-brief.md in my workspace"

Your agent writes the file. You'll find it at ~/.openclaw/workspace/project-brief.md.

File read:

"I put a file called notes.txt in my workspace. Can you organize it into sections?"

First drop a file at ~/.openclaw/workspace/notes.txt, then send the message.

Security model

OpenClaw follows a permission model:

• Tools not in your config cannot be used
• read and write only have access to ~/.openclaw/workspace/ by default
• exec only runs commands on your allowlist
• You can add "confirm": true to any tool to require your approval before each use

Example with confirmation:

{
  "tools": {
    "exec": {
      "enabled": true,
      "allowlist": ["git commit", "git push"],
      "confirm": true
    }
  }
}

With "confirm": true, OpenClaw will ask you to approve before running any exec command. You'll see a prompt in your terminal and must type y to proceed.

Goal complete

You now have a personal assistant that:

• Lives on your messaging app
• Knows how you communicate and what you're working on
• Can fetch URLs, search the web, and manage files
• Remembers key facts between sessions

This is a genuinely useful daily driver. Use it for a week and you'll naturally find new things to add.

Where to go next

• Level Up: The Memory System → — build a knowledge base that grows over time
• Level Up: Skills & ClawHub → — connect Gmail, Calendar, Notion, and more
• Level Up: Security Fundamentals → — if you're sharing the agent with others
• Pick another goal →