🚀 Getting Started
Install the Agent OS VS Code Extension and create your first safe AI agent in under 5 minutes.
Installation
Option 1: VS Code Marketplace (Recommended)
1 Open VS Code
2 Press Ctrl+Shift+X to open Extensions
3 Search for "Agent OS"
4 Click Install
Option 2: Command Line
code --install-extension agent-os.agent-os-vscode
Verify Installation
Press Ctrl+Shift+P and type Agent OS. You should see available commands:
- Agent OS: Getting Started
- Agent OS: Open Policy Editor
- Agent OS: Open Workflow Designer
- And 15 more...
Interactive Onboarding
The extension includes an interactive onboarding experience to get you productive quickly.
Launch Onboarding
Ctrl+Shift+P → "Agent OS: Getting Started"
Onboarding Steps
| Step | Description | Action |
|---|---|---|
| 1. Install Extension | Verify extension is installed | ✅ Auto-completed |
| 2. Configure Policies | Set up safety rules | Opens Policy Editor |
| 3. Create First Agent | Build your first agent | Creates template project |
| 4. Run Safety Test | Verify policy enforcement | Runs validation |
Your First Agent in 5 Minutes
1 Open Getting Started panel
Press Ctrl+Shift+P → type "Agent OS: Getting Started"
2 Click "Create First Agent"
Choose a template (e.g., "Data Processor")
3 Review generated files
The extension creates:
agent.py- Main agent codepolicy.yaml- Safety policyREADME.md- Documentation
4 Run your agent
python agent.py
💡 Tip: Your agent is now protected by kernel-level safety. Any action that violates the policy will be automatically blocked.
Generated Agent Code
"""
Data Processor Agent
Generated by Agent OS VS Code Extension
"""
from agent_os import KernelSpace, Policy
from agent_os.tools import create_safe_toolkit
# Initialize kernel with safety guarantees
kernel = KernelSpace(policy="strict")
toolkit = create_safe_toolkit("standard")
@kernel.agent
async def data_processor(task: str):
"""Process data with full safety guarantees"""
# Read input (policy-checked)
data = await toolkit.file.read("/data/input.csv")
# Process with LLM (rate-limited)
result = await toolkit.llm.call(
model="gpt-4",
prompt=f"Process this data: {data[:1000]}"
)
# Write output (restricted to /tmp/)
await toolkit.file.write("/tmp/output.json", result)
return {"status": "success", "output": "/tmp/output.json"}
if __name__ == "__main__":
import asyncio
result = asyncio.run(kernel.execute(data_processor, "Process sales data"))
print(result)
Next Steps
Now that you have the extension installed, explore these features:
- 📜 Policy Editor - Create and manage safety policies
- 🔧 Workflow Designer - Build workflows visually
- 🔒 Security Diagnostics - Real-time vulnerability detection