LLM Apps in Production (RAG + Vector DB + Caching) · Lesson

Monitoring Costs and Latency

Set up tools and practices to track LLM API costs and application latency, enabling continuous optimization.

Monitoring Costs and Latency is a free LLM Apps in Production (RAG + Vector DB + Caching) lesson on CoddyKit — lesson 3 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the LLM Apps in Production (RAG + Vector DB + Caching) learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.

Crucial for LLM App Health

Deploying Large Language Model (LLM) applications to production comes with unique challenges. Two critical aspects to continuously monitor are operational costs and application latency.

Monitoring helps you ensure your LLM app runs smoothly, efficiently, and within budget, delivering a great user experience.

Understanding LLM API Costs

Most LLM providers charge based on token usage. A token is a piece of a word, like 'hel' or 'lo'. You typically pay for:

Input Tokens: The text you send to the LLM (your prompt and context).
Output Tokens: The text the LLM generates as its response.

Prices vary by model and token type, so tracking usage is key to managing expenses.

Provider Dashboards for Costs

The simplest way to start tracking LLM costs is by using the dashboards provided by your LLM API vendor (e.g., OpenAI, Anthropic). These dashboards usually offer:

An overview of your total spending.
Breakdowns of usage by specific models.
Historical data and trend analysis.

They provide a convenient, high-level view of your expenditure.

Programmatic Cost Tracking

For more granular control and integration into your own systems, you can log token usage directly from your application. LLM API responses often include detailed token counts. Here's a Python example:

import openai

# This client would be initialized with your API key
# client = openai.OpenAI(api_key="YOUR_OPENAI_API_KEY")

def get_llm_response_with_cost(prompt):
    try:
        # Simulate an LLM call without actual API key setup
        # In a real app, 'client.chat.completions.create(...)' would be used
        response_mock = type('obj', (object,), {
            'choices': [type('obj', (object,), {'message': type('obj', (object,), {'content': 'The capital of France is Paris.'})})],
            'usage': type('obj', (object,), {
                'prompt_tokens': 10,
                'completion_tokens': 5,
                'total_tokens': 15
            })
        })()
        
        usage = response_mock.usage # In real code: response.usage
        print(f"Prompt Tokens: {usage.prompt_tokens}")
        print(f"Completion Tokens: {usage.completion_tokens}")
        print(f"Total Tokens: {usage.total_tokens}")
        return response_mock.choices[0].message.content # In real code: response.choices[0].message.content
    except Exception as e:
        print(f"Error: {e}")
        return "Error generating response."

if __name__ == "__main__":
    print("--- LLM Cost Logging Demo --- ")
    get_llm_response_with_cost("What is the capital of France?")

Understanding Latency in RAG

Latency refers to the delay between sending a request and receiving a response. For a Retrieval Augmented Generation (RAG) application, this isn't just the LLM call; it includes several stages:

Time to retrieve documents from your vector database.
The actual LLM API call duration.
Any preprocessing or postprocessing steps.

High latency can lead to a frustratingly slow user experience.

Measuring Latency in Your App

To optimize your RAG system's performance, you need to identify where delays are occurring. This means measuring the time taken for each critical component of your pipeline:

Data ingestion and chunking.
Embedding generation.
Vector database queries.
LLM API calls.

Python's time module is a simple yet effective tool for this.

Practical Latency Logging

Let's extend our previous example to measure the duration of an LLM call. This is often the most significant contributor to overall RAG latency:

import openai
import time

# This client would be initialized with your API key
# client = openai.OpenAI(api_key="YOUR_OPENAI_API_KEY")

def get_llm_response_timed(prompt):
    start_time = time.time()
    try:
        # Simulate an LLM call without actual API key setup
        # In a real app, 'client.chat.completions.create(...)' would be used
        # Simulate a network delay
        time.sleep(0.5)
        response_mock = type('obj', (object,), {
            'choices': [type('obj', (object,), {'message': type('obj', (object,), {'content': 'Once upon a time, there was a brave knight.'})})],
        })()
        
        end_time = time.time()
        duration = end_time - start_time
        print(f"LLM Call Duration: {duration:.2f} seconds")
        return response_mock.choices[0].message.content # In real code: response.choices[0].message.content
    except Exception as e:
        print(f"Error: {e}")
        return "Error generating response."

if __name__ == "__main__":
    print("--- LLM Latency Logging Demo --- ")
    get_llm_response_timed("Tell me a short story about a brave knight.")

Centralizing Metrics & Tools

For a holistic view of your application's health, it's best to centralize your logs and metrics using dedicated monitoring tools. Popular choices include:

Prometheus: Excellent for collecting and storing time-series data (metrics).
Grafana: For building powerful, customizable dashboards and visualizations.
Datadog / New Relic: All-in-one observability platforms that combine metrics, logs, and traces.

These platforms help you visualize trends and quickly pinpoint issues.

Setting Up Proactive Alerts

While monitoring helps you understand what's happening, alerting ensures you're notified immediately when something goes wrong. Configure alerts to trigger if:

Your monthly LLM API costs exceed a predefined budget.
The average response latency for your RAG system spikes unexpectedly.
Error rates for LLM calls or retrieval increase significantly.

Proactive alerts enable you to address problems before they negatively impact users or your budget.

Quick Check: Monitoring Costs

You've learned about tracking LLM costs and latency. Let's test your understanding of why monitoring token usage is so important.

Recap: Monitor for Success

Monitoring costs and latency is absolutely vital for any production LLM application. By programmatically tracking token usage and timing key operations, you gain crucial insights to optimize your system's performance and manage budgets effectively.

Integrating with observability platforms and setting up proactive alerts ensures your RAG system remains efficient, cost-effective, and provides a reliable user experience.

Frequently asked questions

Is the “Monitoring Costs and Latency” lesson free?

Yes — the full text of “Monitoring Costs and Latency” is free to read here on the web, and the LLM Apps in Production (RAG + Vector DB + Caching) course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the LLM Apps in Production (RAG + Vector DB + Caching) course, upgrade to CoddyKit PRO.

What will I learn in “Monitoring Costs and Latency”?

Set up tools and practices to track LLM API costs and application latency, enabling continuous optimization. You practise LLM Apps in Production (RAG + Vector DB + Caching) with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.

Do I need any experience to start LLM Apps in Production (RAG + Vector DB + Caching)?

No prior experience is required. LLM Apps in Production (RAG + Vector DB + Caching) on CoddyKit is structured for beginners through advanced learners; this is — lesson 3 of 4, so you can start here or from the beginning and move at your own pace.

How long does the “Monitoring Costs and Latency” lesson take?

Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.

Can I write and run code in this LLM Apps in Production (RAG + Vector DB + Caching) lesson?

Yes. Every LLM Apps in Production (RAG + Vector DB + Caching) lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.

All lessons in this course

← Back to LLM Apps in Production (RAG + Vector DB + Caching)