Best Practices #

Recommended patterns for building reliable applications with the Nellie API.

Security #

API Key Management #

✅ DO #

# Use environment variables
import os
api_key = os.environ.get("NELLIE_API_KEY")

# Or use a secrets manager
from aws_secretsmanager import get_secret
api_key = get_secret("nellie/api-key")

❌ DON’T #

# Never hardcode API keys
api_key = "nel_abc123..."  # BAD!

# Never commit to version control
# .env files should be in .gitignore

Key Rotation #

Rotate API keys regularly, especially:

When team members leave
After potential exposure
As part of regular security hygiene

# Zero-downtime key rotation
# 1. Create new key in API dashboard
# 2. Update application config
new_key = os.environ.get("NELLIE_API_KEY_NEW")
client = Nellie(api_key=new_key)

# 3. Deploy and verify
# 4. Revoke old key in API dashboard

Webhook Security #

Always verify webhook signatures:

from nellie_api import Webhook, WebhookSignatureError

@app.route("/webhook", methods=["POST"])
def webhook():
    try:
        event = Webhook.construct_event(
            request.data,
            request.headers.get("X-Nellie-Signature"),
            WEBHOOK_SECRET
        )
    except WebhookSignatureError:
        # Log the attempt for security monitoring
        logger.warning(f"Invalid webhook signature from {request.remote_addr}")
        return "Invalid signature", 400
    
    # Process verified event
    handle_event(event)
    return "OK", 200

Server-Side Only #

Never expose your API key in client-side code:

// ❌ BAD - Client-side JavaScript
const response = await fetch('https://api.nelliewriter.com/v1/book', {
  headers: { 'X-API-Key': 'nel_abc123...' }  // Exposed to users!
});

// ✅ GOOD - Call your own backend
const response = await fetch('/api/generate-book', {
  method: 'POST',
  body: JSON.stringify({ prompt: 'A mystery novel' })
});

Your backend then calls Nellie:

# Backend API route
@app.route("/api/generate-book", methods=["POST"])
def generate_story():
    client = Nellie(api_key=os.environ["NELLIE_API_KEY"])
    book = client.books.create(prompt=request.json["prompt"])
    return jsonify({"requestId": book.request_id})

Error Handling #

Implement Retry Logic #

from nellie_api import Nellie, RateLimitError, APIError
import time
from functools import wraps

def with_retry(max_attempts=3, backoff_factor=2):
    ""Decorator for retry logic with exponential backoff.""
    def decorator(func):
        @wraps(func)
        def wrapper(*args, **kwargs):
            last_exception = None
            
            for attempt in range(max_attempts):
                try:
                    return func(*args, **kwargs)
                except RateLimitError as e:
                    last_exception = e
                    wait_time = backoff_factor ** attempt * 30
                    print(f"Rate limited. Waiting {wait_time}s...")
                    time.sleep(wait_time)
                except APIError as e:
                    if e.status_code >= 500:
                        last_exception = e
                        wait_time = backoff_factor ** attempt * 10
                        time.sleep(wait_time)
                    else:
                        raise  # Don't retry client errors
            
            raise last_exception
        return wrapper
    return decorator

@with_retry(max_attempts=3)
def create_book(prompt: str):
    client = Nellie()
    return client.books.create(prompt=prompt)

Handle All Error Types #

from nellie_api import (
    NellieError,
    AuthenticationError,
    RateLimitError,
    APIError
)

def safe_generate(prompt: str) -> dict:
    ""Generate with comprehensive error handling.""
    try:
        client = Nellie()
        book = client.books.create(prompt=prompt)
        return {"success": True, "request_id": book.request_id}
        
    except AuthenticationError:
        return {
            "success": False,
            "error": "auth_error",
            "message": "Invalid API key. Check your configuration."
        }
        
    except RateLimitError:
        return {
            "success": False,
            "error": "rate_limit",
            "message": "Too many requests. Please try again later."
        }
        
    except APIError as e:
        return {
            "success": False,
            "error": "api_error",
            "message": str(e),
            "status_code": e.status_code
        }
        
    except NellieError as e:
        return {
            "success": False,
            "error": "unknown",
            "message": str(e)
        }

Graceful Degradation #

def generate_content(prompt: str) -> str:
    ""Generate content with fallback behavior.""
    try:
        client = Nellie()
        book = client.books.create(prompt=prompt)
        result = client.books.wait_for_completion(book.request_id)
        
        if result.is_successful():
            return result.result_url
        else:
            # Fallback: notify user of failure
            return notify_user_of_delay(book.request_id)
            
    except NellieError:
        # Fallback: queue for later processing
        return queue_for_retry(prompt)

Performance #

Use Webhooks Over Polling #

Webhooks are more efficient and provide faster notifications:

# ✅ Preferred: Webhook-based flow
book = client.books.create(
    prompt="A mystery novel",
    webhook_url="https://myapp.com/webhooks/nellie"
)
# Your server receives a POST when complete

# ❌ Less efficient: Polling
book = client.books.create(prompt="A mystery novel")
while True:
    status = client.books.retrieve(book.request_id)
    if status.is_complete():
        break
    time.sleep(120)  # Wasted API calls

Respect Rate Limits #

import time
from collections import deque
from threading import Lock

class RateLimiter:
    ""Simple rate limiter for API calls.""
    
    def __init__(self, max_calls: int, period: float):
        self.max_calls = max_calls
        self.period = period
        self.calls = deque()
        self.lock = Lock()
    
    def acquire(self):
        with self.lock:
            now = time.time()
            
            # Remove old calls
            while self.calls and self.calls[0] < now - self.period:
                self.calls.popleft()
            
            # Wait if at limit
            if len(self.calls) >= self.max_calls:
                sleep_time = self.calls[0] + self.period - now
                time.sleep(sleep_time)
                return self.acquire()
            
            self.calls.append(now)

# Usage
limiter = RateLimiter(max_calls=1, period=6)  # 1 request per 6 seconds

def create_book_limited(prompt: str):
    limiter.acquire()
    return client.books.create(prompt=prompt)

Cache Configuration Data #

from functools import lru_cache
from datetime import datetime, timedelta

@lru_cache(maxsize=1)
def get_cached_config():
    ""Cache configuration for 1 hour.""
    return client.get_configuration()

# Or with time-based expiration
class ConfigCache:
    def __init__(self, ttl_seconds=3600):
        self.ttl = ttl_seconds
        self.config = None
        self.expires_at = None
    
    def get(self):
        now = datetime.now()
        if self.config is None or now > self.expires_at:
            self.config = client.get_configuration()
            self.expires_at = now + timedelta(seconds=self.ttl)
        return self.config

config_cache = ConfigCache()
styles = config_cache.get().styles

Reliability #

Idempotent Webhook Handlers #

Handle duplicate webhook deliveries gracefully:

from functools import lru_cache

# Track processed events
processed_events = set()

@app.route("/webhook", methods=["POST"])
def webhook():
    event = Webhook.construct_event(...)
    
    # Check if already processed
    if event.request_id in processed_events:
        return "Already processed", 200
    
    # Process the event
    process_book(event)
    
    # Mark as processed
    processed_events.add(event.request_id)
    
    return "OK", 200

For production, use a database:

def handle_webhook(event):
    # Use database with unique constraint
    try:
        db.execute(
            "INSERT INTO processed_webhooks (request_id) VALUES (%s)",
            (event.request_id,)
        )
    except UniqueViolation:
        return  # Already processed
    
    # Process the event
    process_book(event)

Timeout Handling #

from nellie_api import Nellie

client = Nellie(
    api_key="nel_...",
    timeout=60  # Request timeout
)

# For long-running operations
try:
    result = client.books.wait_for_completion(
        request_id,
        timeout=7200,  # 2 hour max wait
        poll_interval=120
    )
except TimeoutError:
    # Handle timeout gracefully
    logger.warning(f"Book {request_id} timed out")
    notify_user_of_delay(request_id)

Health Checks #

Monitor API availability:

import requests

def check_nellie_health() -> bool:
    ""Check if Nellie API is responsive.""
    try:
        response = requests.get(
            "https://api.nelliewriter.com/v1/configuration",
            timeout=10
        )
        return response.status_code == 200
    except requests.RequestException:
        return False

# Use in health check endpoint
@app.route("/health")
def health():
    checks = {
        "database": check_database(),
        "nellie_api": check_nellie_health(),
    }
    
    all_healthy = all(checks.values())
    return jsonify(checks), 200 if all_healthy else 503

Architecture #

Asynchronous Processing #

Don’t block web requests on book generation:

# ✅ Good: Return immediately, process async
@app.route("/generate", methods=["POST"])
def generate():
    prompt = request.json["prompt"]
    
    # Queue the job
    job_id = queue_generation_job(prompt)
    
    return jsonify({
        "status": "queued",
        "job_id": job_id,
        "message": "Generation started."
    }), 202

# Background worker processes the queue
def worker():
    while True:
        job = get_next_job()
        client = Nellie()
        
        book = client.books.create(
            prompt=job.prompt,
            webhook_url=f"{BASE_URL}/internal/webhook/{job.id}"
        )
        
        update_job_status(job.id, "processing", book.request_id)

Separate Concerns #

# services/nellie_service.py
class NellieService:
    ""Encapsulate all Nellie API interactions.""
    
    def __init__(self):
        self.client = Nellie()
    
    def start_generation(self, prompt: str, **kwargs) -> str:
        ""Start a book generation and return the request ID.""
        book = self.client.books.create(prompt=prompt, **kwargs)
        return book.request_id
    
    def get_status(self, request_id: str) -> dict:
        ""Get generation status.""
        status = self.client.books.retrieve(request_id)
        return {
            "status": status.status,
            "progress": status.progress,
            "result_url": status.result_url,
            "error": status.error
        }
    
    def get_available_credits(self) -> int:
        ""Calculate available credits.""
        usage = self.client.get_usage()
        return CREDIT_LIMIT - usage.total_credits_used

# Use in routes
nellie = NellieService()

@app.route("/books", methods=["POST"])
def create_book():
    request_id = nellie.start_generation(request.json["prompt"])
    return jsonify({"request_id": request_id}), 202

Monitoring #

Log Important Events #

import logging

logger = logging.getLogger("nellie")

def generate_book(prompt: str):
    logger.info(f"Starting generation", extra={"prompt_length": len(prompt)})
    
    try:
        book = client.books.create(prompt=prompt)
        logger.info(f"Generation started", extra={
            "request_id": book.request_id
        })
        
        result = client.books.wait_for_completion(book.request_id)
        
        if result.is_successful():
            logger.info("Generation completed", extra={
                "request_id": result.request_id,
                "credits_used": result.credits_used
            })
        else:
            logger.error("Generation failed", extra={
                "request_id": result.request_id,
                "error": result.error
            })
            
    except Exception as e:
        logger.exception("Unexpected error during generation")
        raise

Track Metrics #

from prometheus_client import Counter, Histogram

# Define metrics
stories_created = Counter(
    'nellie_stories_created_total',
    'Total stories created',
    ['status']
)

generation_duration = Histogram(
    'nellie_generation_duration_seconds',
    'Time to complete generation'
)

credits_used = Counter(
    'nellie_credits_used_total',
    'Total credits consumed'
)

# Use in code
def generate_with_metrics(prompt: str):
    with generation_duration.time():
        book = client.books.create(prompt=prompt)
        result = client.books.wait_for_completion(book.request_id)
    
    stories_created.labels(status=result.status).inc()
    
    if result.is_successful():
        credits_used.inc(result.credits_used)
    
    return result

Alert on Issues #

def handle_webhook(event):
    if event.status == "failed":
        # Send alert
        send_alert(
            level="warning",
            title="Book Generation Failed",
            message=f"Request {event.request_id} failed: {event.error}"
        )
    
    # Track failure rate
    if should_alert_on_failure_rate():
        send_alert(
            level="critical",
            title="High Failure Rate Detected",
            message="Book generation failure rate exceeded threshold"
        )

Cost Management #

Monitor Credit Usage #

def check_credit_budget():
    ""Alert if credit usage exceeds budget.""
    usage = client.get_usage()
    
    MONTHLY_BUDGET = 50000
    ALERT_THRESHOLD = 0.8
    
    if usage.total_credits_used > MONTHLY_BUDGET * ALERT_THRESHOLD:
        send_alert(
            level="warning",
            title="Credit Budget Alert",
            message=f"Used {usage.total_credits_used}/{MONTHLY_BUDGET} credits"
        )

Pre-flight Credit Check #

def can_afford_generation(model: str = "2.0") -> bool:
    ""Check if we have enough credits.""
    COSTS = {"2.0": 250, "3.0": 500}
    BUFFER = 100  # Keep some reserve
    
    usage = client.get_usage()
    available = CREDIT_LIMIT - usage.total_credits_used
    
    return available >= COSTS.get(model, 500) + BUFFER

Summary Checklist #

Before Going to Production #

[ ] API key stored in environment variable or secrets manager
[ ] Webhook signatures are verified
[ ] Error handling covers all exception types
[ ] Retry logic implemented with exponential backoff
[ ] Rate limits respected
[ ] Webhook handlers are idempotent
[ ] Logging captures important events
[ ] Monitoring and alerts configured
[ ] Credit usage tracking in place
[ ] Health checks implemented