⚡ Quickstart

Get Auditail protecting your AI agents in under 5 minutes.

---

Step 1 — Install the SDK

npm

npm install auditail-sdk

yarn

yarn add auditail-sdk

Python (pip)

pip install auditail

---

Step 2 — Get an API Key

👉 Request your API key here

You'll get your key within 24 hours. During the private beta, all keys are free.

---

Step 3 — Initialize the Client

TypeScript

import { Auditail } from 'auditail-sdk';

const aegis = new Auditail({
  apiKey: process.env.AUDITAIL_API_KEY!,
  environment: 'production', // shows in the dashboard
});

Python

import os
from auditail import Auditail

aegis = Auditail(
    api_key=os.environ["AUDITAIL_API_KEY"],
    environment="production",
)

---

Step 4 — Add Guardrail Rules

TypeScript

aegis
  .addRule(Auditail.rules.noExternalEmail())
  .addRule(Auditail.rules.noDataDeletion())
  .addRule(Auditail.rules.noExternalApiCalls());

Python

from auditail import rules

aegis.add_rules([
    rules.no_external_email(),
    rules.no_data_deletion(),
    rules.no_external_api_calls(),
])

---

Step 5 — Wrap Every Agent Action

This is the core pattern. Before your AI agent executes any action, call aegis.guard():

TypeScript

const result = await aegis.guard({
  action: 'send_email',
  payload: {
    to: 'client@example.com',
    subject: 'Q3 Report',
    body: 'Please find the report attached.',
  },
  agentId: 'financial-report-bot',
  userId: req.user.id, // optional, for user-level audit trails
});

if (result.approved) {
  await emailService.send(result.payload);
} else {
  console.log(`Blocked [${result.matchedRule}]: ${result.reason}`);
  // result.auditId is already logged — reference it in your incident report
}

Python

result = aegis.guard(
    action="send_email",
    payload={
        "to": "client@example.com",
        "subject": "Q3 Report",
        "body": "Please find the report attached.",
    },
    agent_id="financial-report-bot",
    user_id=request.user_id,  # optional
)

if result.approved:
    email_service.send(**result.payload)
else:
    print(f"Blocked [{result.matched_rule}]: {result.reason}")

---

What You Get

Every guard() call returns:

{
  "approved": false,
  "auditId": "aeg_1a2b3c_x9z4k7",
  "reason": "Blocked: sending email to an external domain requires human approval.",
  "matchedRule": "no-external-email",
  "timestamp": "2025-01-15T10:30:00.000Z",
  "latencyMs": 2
}

And every event is logged to the Auditail dashboard with:

• Full payload
• The agent ID and user ID
• Which rule fired and why
• Environment label

---

Next Steps

• Guardrails → — Write custom rules for your specific use case
• Audit Trail → — Understand what gets logged and how to query it
• LangChain Integration → — Drop-in wrapper for LangChain tools
• Human Approval → — Pause on risky actions and require a human sign-off

tip

The preview SDK runs fully locally — guardrail evaluation and in-memory logging work out of the box. Only the dashboard and persistent log shipping require a beta API key.