Field Notes/Getting Started
🔍 Discovery

Why Your OpenClaw Agent Needs Separate Accounts (Security Setup)

2 min read

Your agent doesn't need your passwords. Give it dedicated accounts — separate email, API keys, cloud access. Everything isolated. Everything revocable.

Treat Your Agent Like a New Hire

You wouldn't hand a new employee your personal Gmail password. You'd create them a company account. Your agent is no different.

My agent Arun has arun@[mydomain].com, its own Drive service account, its own API keys. If I need to revoke access, I turn off his credentials — not mine.

What to Create

Here's a typical agent account setup:

  • Email: Dedicated address on your domain (or free Gmail for testing)
  • Storage: Google Drive service account scoped to one shared folder
  • API Keys: Separate keys for each service (OpenAI, Anthropic, etc.)
  • GitHub: Fine-grained PAT scoped to specific repos
  • Cloud: IAM roles with minimal permissions

The pattern: isolated and revocable. Your agent gets exactly what it needs, nothing more.

Why This Matters for Autonomous Agents

Once your agent starts acting autonomously — sending emails, hitting APIs, writing to databases — you need two things:

  • A clear paper trail — see exactly what your agent did
  • A kill switch — that doesn't kill your access too

Some people go further: creating the agent as a full Google Workspace user with its own login and 2FA. A bit paranoid, but it's clean architecture.

The Bonus: Better Agent Design

Separate accounts force you to think about permissions explicitly. Read-only here? Write access there? Can it send emails or just draft them?

Fewer permissions = less surface area for mistakes. An agent that can only draft emails gives you a chance to review before anything goes out.

Start Simple

Even a Gmail and a separate API key is enough to start. Build good habits now — scale complexity as your agent grows.

Your future self — the one debugging why an agent sent weird emails from your personal account at 3 AM — will thank you.