Step-by-step: Integrating ElevenLabs text-to-speech into my Notion workflow.

Everyone's showing off their cool Notion+ElevenLabs automations. Nobody's talking about the IAM mess and data exposure they're creating.

Here's the secure way to do it. The core problem: your ElevenLabs API key in a plaintext environment variable in some random automation platform is a liability.

**Step 1: Isolate the API Call**
Don't call ElevenLabs directly from Notion/Make/Zapier. Use an AWS Lambda function as a proxy. This keeps your key off third-party servers.

**Step 2: Least Privilege IAM**
The execution role for the Lambda should have *zero* permissions beyond logging. It doesn't need S3, DynamoDB, or network access.

**Step 3: Secure Key Storage**
Store the ElevenLabs API key in AWS Secrets Manager. Lambda fetches it at runtime. Rotate it regularly.

Basic Lambda proxy structure (Python):

```python
import json
import requests
import os
import boto3
from botocore.exceptions import ClientError

def lambda_handler(event, context):
# Get secret
secret_name = "elevenlabs/api-key"
client = boto3.client('secretsmanager')
try:
secret_response = client.get_secret_value(SecretId=secret_name)
api_key = secret_response['SecretString']
except ClientError as e:
raise e

# Get text from event (parsed from Notion webhook)
text_to_speak = event.get('text')

# Call ElevenLabs
headers = {
"Accept": "audio/mpeg",
"Content-Type": "application/json",
"xi-api-key": api_key
}
response = requests.post(
f"https://api.elevenlabs.io/v1/text-to-speech/YOUR_VOICE_ID",
json={"text": text_to_speak, "voice_settings": {"stability": 0.5, "similarity_boost": 0.5}},
headers=headers
)

# Return audio stream or store in a *private* S3 bucket with a presigned URL
return {
'statusCode': 200,
'body': json.dumps({'audio_url': '...'})
}
```

**Step 4: Lock Down the Trigger**
If using API Gateway, use a strong API key and request validation. Monitor logs for unusual call volume.

This adds complexity, but it prevents a leaked key from being used to drain your ElevenLabs credits or exfiltrate sensitive text processed through the pipeline.