> ## Documentation Index
> Fetch the complete documentation index at: https://docs.polaxis.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Quick Start — Add AI Agent Security in 5 Minutes

> Install Polaxis and add runtime security, policy enforcement, and governance to any AI agent in under 5 minutes. Works with OpenAI, LangChain, CrewAI, and any custom agent.

# Quick Start

Get Polaxis running in under 5 minutes.

## Install

```bash theme={null}
pip install polaxis
```

## Get your API key

1. Sign up at [polaxis.io](https://polaxis.io)
2. Go to **Dashboard → Agents → Create Agent**
3. Copy your API key — it starts with `ag_`

## Add to your agent

```python theme={null}
from polaxis import Polaxis

guard = Polaxis(
    api_key="ag_prod_...",
    agent_id="your-agent-id"
)

# Before every tool call:
result = await guard.evaluate("send_email", {
    "to": "customer@example.com",
    "subject": "Invoice",
    "body": "..."
})
# → allow | escalate (block raises BlockedError)
```

That's it. Every tool call your agent makes is now governed.

## What happens next

| Decision   | Meaning              | Action                             |
| ---------- | -------------------- | ---------------------------------- |
| `allow`    | Safe to proceed      | Tool call executes                 |
| `block`    | Policy violation     | `BlockedError` raised              |
| `escalate` | Needs human approval | SDK waits, Slack notification sent |

## Full example

```python theme={null}
import asyncio
from polaxis import Polaxis, BlockedError

guard = Polaxis(api_key="ag_prod_...", agent_id="billing-agent")

async def run_agent():
    tools = [
        ("send_email", {"to": "user@co.com", "body": "Your invoice is ready"}),
        ("charge_card", {"amount": 4999, "currency": "usd"}),
        ("delete_records", {"table": "users", "where": "inactive=true"}),
    ]

    for tool_name, tool_input in tools:
        try:
            result = await guard.evaluate(tool_name, tool_input)
            print(f"✓ {tool_name} — {result.decision}")
            # ... execute your tool here
        except BlockedError as e:
            print(f"✗ {tool_name} blocked — {e.reason}")

asyncio.run(run_agent())
```