← Back|AI-AGENTS›Section 1/16

0 of 16 completed

Using APIs inside agents

Intermediate⏱ 10 min read📅 Updated: 2026-02-17

🔌 Introduction – APIs Give Agents Superpowers!

AI Agent brain irukku (LLM), but hands irukkaadhu – real world la enna pannum?

APIs are those hands! 🤲

Without APIs: Agent just text generate pannum

With APIs: Agent email send pannum, data fetch pannum, payments process pannum, bookings confirm pannum!

Agent Alone	Agent + APIs
"Here are flight options"	Actually books the flight ✈️
"You should send an email"	Sends the email 📧
"Current weather might be..."	Gets real-time weather 🌤️
"You could check your calendar"	Reads your actual calendar 📅

APIs transform agents from advisors to executors! 💪

📚 API Fundamentals for Agents

API Types agents commonly use:

Type	Protocol	Format	Example
REST	HTTP	JSON	Most web APIs
GraphQL	HTTP	JSON	GitHub, Shopify
WebSocket	WS	JSON/Binary	Real-time data
gRPC	HTTP/2	Protocol Buffers	Microservices

REST API Basics (most common):

code

GET    /api/flights?from=MAA&to=BLR    → Fetch data
POST   /api/bookings                    → Create data
PUT    /api/bookings/123                → Update data
DELETE /api/bookings/123                → Delete data

HTTP Status Codes agents should handle:

Code	Meaning	Agent Action
200	Success ✅	Process response
400	Bad request	Fix parameters, retry
401	Unauthorized	Refresh auth token
404	Not found	Try alternative
429	Rate limited	Wait and retry
500	Server error	Retry with backoff

🔧 How Agents Use APIs – Function Calling

Modern LLMs function calling feature irukku – agent automatically correct API call pannum!

How it works:

Step 1: Define available tools (APIs)

json

{
  "name": "search_flights",
  "description": "Search available flights between cities",
  "parameters": {
    "from_city": "string (IATA code)",
    "to_city": "string (IATA code)",
    "date": "string (YYYY-MM-DD)",
    "passengers": "integer"
  }
}

Step 2: User gives request

"Book me a flight to Bangalore tomorrow"

Step 3: LLM decides which function to call

json

{
  "function": "search_flights",
  "arguments": {
    "from_city": "MAA",
    "to_city": "BLR",
    "date": "2026-02-18",
    "passengers": 1
  }
}

Step 4: Agent executes the API call

Step 5: Response goes back to LLM for interpretation

This is how ALL modern AI agents work! 🔧

🏗️ API Integration Architecture

🏗️ Architecture Diagram

```
┌─────────────────────────────────────────┐
│           USER REQUEST                  │
│  "What's the weather in Chennai?"       │
└──────────────────┬──────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────┐
│  🧠 LLM REASONING                      │
│  "I need to call weather API"           │
│  → Function: get_weather("Chennai")     │
└──────────────────┬──────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────┐
│  🔧 API EXECUTION LAYER                │
│  ┌─────────────────────────────────┐   │
│  │ 1. Validate parameters          │   │
│  │ 2. Add authentication           │   │
│  │ 3. Make HTTP request            │   │
│  │ 4. Handle errors/retries        │   │
│  │ 5. Parse response               │   │
│  └─────────────────────────────────┘   │
└──────────────────┬──────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────┐
│  🌐 EXTERNAL API                        │
│  api.openweathermap.org/weather         │
│  Response: {"temp": 32, "desc": "sunny"}│
└──────────────────┬──────────────────────┘
                   │
                   ▼
┌─────────────────────────────────────────┐
│  🧠 LLM INTERPRETATION                 │
│  "Chennai la ippo 32°C, sunny day! ☀️"  │
└─────────────────────────────────────────┘
```

🔑 Authentication Methods

APIs access panna authentication venum:

Method	How It Works	Security	Example
API Key	Key in header/query	Basic	OpenWeather
Bearer Token	JWT token in header	Good	Most APIs
OAuth 2.0	Token exchange flow	Strong	Google, GitHub
Basic Auth	Username:password	Weak	Legacy APIs

API Key Example:

code

GET /api/weather?city=Chennai
Headers: {
  "X-API-Key": "your-secret-key-here"
}

Bearer Token Example:

code

GET /api/user/profile
Headers: {
  "Authorization": "Bearer eyJhbGciOi..."
}

Security Best Practices:

🔒 Never hardcode API keys in source code
📁 Use environment variables (.env files)
🔄 Rotate keys periodically
🛡️ Use minimum required permissions
📋 Monitor API key usage for anomalies

🎬 Real Example – Travel Agent APIs

✅ Example

Building a Travel Planning Agent:

code

Agent has access to 5 APIs:

1. 🌤️ Weather API
   GET /weather?city=Goa&date=2026-03-15
   → {"temp": 33, "condition": "sunny", "humidity": 75}

2. ✈️ Flight API  
   GET /flights?from=MAA&to=GOI&date=2026-03-15
   → [{"airline": "IndiGo", "price": 3500, "time": "08:30"}]

3. 🏨 Hotel API
   GET /hotels?city=Goa&checkin=2026-03-15&nights=3
   → [{"name": "Beach Resort", "price": 5000, "rating": 4.5}]

4. 🍽️ Restaurant API
   GET /restaurants?city=Goa&cuisine=seafood
   → [{"name": "Fisherman's Wharf", "rating": 4.8}]

5. 📧 Email API
   POST /send-email {to, subject, body}
   → {"status": "sent"}

User: "Plan a Goa trip for March 15, 3 nights"

Agent calls 4 APIs in parallel, then sends summary via Email API!

All automated, all in under 30 seconds! ⚡

⚠️ Error Handling for APIs

APIs WILL fail. Plan for it!

Error Handling Strategy:

code

function callAPI(endpoint, params, maxRetries = 3) {
  for (attempt = 1; attempt <= maxRetries; attempt++) {
    try {
      response = http.get(endpoint, params)
      
      if (response.status === 200) return response.data
      if (response.status === 429) {
        wait(attempt * 2 seconds)  // exponential backoff
        continue
      }
      if (response.status === 401) {
        refreshToken()
        continue
      }
      if (response.status >= 500) {
        wait(attempt * 1 second)
        continue
      }
      
      throw Error("Unexpected status: " + response.status)
    } catch (networkError) {
      if (attempt === maxRetries) throw networkError
      wait(attempt * 2 seconds)
    }
  }
}

Error Recovery Matrix:

Error	Retryable?	Strategy
Timeout	Yes	Retry 3x with backoff
Rate limit (429)	Yes	Wait and retry
Auth error (401)	Once	Refresh token, retry
Bad request (400)	No	Fix params or skip
Server error (500)	Yes	Retry 3x
Not found (404)	No	Use alternative

🚀 Popular APIs for Agents

Must-know APIs for building agents:

Category	API	Free Tier	Use Case
🌤️ Weather	OpenWeatherMap	1000 calls/day	Weather data
📧 Email	SendGrid	100 emails/day	Send emails
🔍 Search	SerpAPI	100 searches/mo	Web search
📅 Calendar	Google Calendar	Free	Schedule management
💬 Messaging	Twilio	Trial credits	SMS/WhatsApp
📊 Data	Alpha Vantage	25 calls/day	Stock data
🗺️ Maps	Google Maps	$200/mo free	Location services
📰 News	NewsAPI	100 requests/day	News articles
🐙 Code	GitHub API	5000 req/hr	Repository management
🧠 AI	OpenAI API	Pay-per-use	LLM capabilities

Pro tip: Start with free tiers. Upgrade only when you need more! 💰

🔄 API Rate Limiting & Optimization

Rate limits handle panradhu critical!

Common Rate Limits:

API	Limit	Period
OpenAI	500 RPM	Per minute
Google Maps	50 QPS	Per second
Twitter	300 req	Per 15 min
GitHub	5000 req	Per hour

Optimization Strategies:

Caching 💾

code

cache = {}
if query in cache and cache_age < 1_hour:
    return cache[query]  // No API call needed!
else:
    result = call_api(query)
    cache[query] = result
    return result

Batching 📦

10 individual API calls → 1 batch API call
Many APIs support batch endpoints

Parallel Requests ⚡

Independent API calls simultaneously execute pannunga
Don't wait for weather API to finish before calling flight API

Request Deduplication 🔄

Same query within 5 minutes? Return cached result

Impact:

Strategy	API Calls Saved
Caching	40-60%
Batching	50-80%
Parallel	0% (but faster!)
Dedup	10-30%

🧪 Try It – Build an API-Using Agent

📋 Copy-Paste Prompt

```
You are an AI Agent with these API tools:

TOOLS:
1. get_weather(city, date) → {temp, condition, humidity}
2. search_news(topic, limit) → [{title, summary, url}]
3. send_email(to, subject, body) → {status}
4. translate(text, from_lang, to_lang) → {translated_text}

USER REQUEST: "Check tomorrow's weather in Mumbai, find 
top 3 AI news articles, translate the headlines to Tamil, 
and email everything to boss@company.com"

Execute step by step:
1. Show which API you'll call and why
2. Show the API call with parameters
3. Show simulated response
4. Show how you combine results
5. Show the final email content

Think about: Can any calls be made in parallel?
```

Practice API orchestration skills! 🔧

⚠️ API Security Checklist

⚠️ Warning

Before integrating any API:

- 🔑 API keys in .env – Never in source code

- 🔒 HTTPS only – No HTTP API calls

- 🛡️ Input validation – Sanitize before sending to API

- 📋 Scope limitation – Request minimum permissions

- 💰 Spending limits – Set budget alerts on paid APIs

- 📊 Usage monitoring – Track calls, detect anomalies

- 🔄 Key rotation – Change keys every 90 days

- 🗑️ Data cleanup – Don't store unnecessary API responses

One leaked API key = potential thousands in charges! 💸

💡 API Integration Best Practices

💡 Tip

1. Wrapper functions – Don't call APIs directly, wrap them

2. Retry logic – Built-in for every API call

3. Timeout handling – Max 30s per API call

4. Response validation – Check response format before using

5. Fallback APIs – Primary fails? Use backup provider

6. Documentation – Document every API your agent uses

7. Version pinning – Use specific API versions, not "latest"

8. Mock testing – Test without hitting real APIs

📝 Summary

Key Takeaways:

✅ APIs give agents real-world action capability

✅ Function calling: LLM decides which API to call and with what params

✅ Authentication: API Key, Bearer Token, OAuth 2.0

✅ Error handling: Retry, backoff, fallback, human escalation

✅ Optimization: Cache, batch, parallel, deduplicate

✅ Security: .env files, HTTPS, input validation, spending limits

✅ Start with free tier APIs, upgrade when needed

Next article la Multi-Agent Architecture paapom – enterprise-grade agent systems! 🏢

🏁 🎮 Mini Challenge

Challenge: Build Agent-API Integration

Agent-ku external API integrate panni hands-on:

Scenario: Travel booking agent – flights + hotels search pannum

Step 1: Define APIs (3 mins)

Identify 3 APIs:

Flight Search API: GET /api/flights?from=MAA&to=BLR&date=2026-02-25
Hotel Search API: GET /api/hotels?city=BLR&checkin=2026-02-25
Weather API: GET /api/weather?city=BLR

Step 2: Create Tool Definitions (4 mins)

Each API-ku tool specification:

json

{
  "name": "search_flights",
  "description": "Search available flights",
  "parameters": {
    "from_city": "string (IATA code)",
    "to_city": "string (IATA code)",
    "date": "string (YYYY-MM-DD)",
    "passengers": "integer"
  }
}

Step 3: Function Calling Setup (3 mins)

LLM-na tools define panni, LLM itself decide panna which tool call pannum

Example: User "Book flight to Bangalore next week"

LLM: "I need to call search_flights"
Params: {from_city: "MAA", to_city: "BLR", date: "2026-03-02", passengers: 1}

Step 4: Error Handling (3 mins)

Each API-ku error cases:

API timeout? → Retry
Auth fail? → Refresh token
Rate limit? → Wait and retry
Invalid params? → Inform LLM, refine

Step 5: Test (2 mins)

Test workflow:

code

User: "Show me flights to Bangalore and weather there"
Agent calls: search_flights() + weather_api()
Results combine aagum
Output: "Found 5 flights, weather is sunny"

API integration complete! 🔌

💼 Interview Questions

Q1: AI Agent-ku APIs edhuku important?

A: APIs = agent-oda hands! Without APIs, agent just text generate pannum. APIs irundha real actions execute pannum – booking, payment, data fetch, etc. Agent brain becomes executor!

Q2: Function calling vs direct API calls – enna difference?

A: Function calling: LLM decides which function call pannanum, automatically! No manual code needed. Agent directly API hit pannum vs. LLM saying "call this function with these params" – abstraction layer!

Q3: API error handling agent perspective?

Timeout: Retry with exponential backoff
Auth fail: Refresh token, retry
Rate limit: Queue, wait, retry
Invalid params: Request better params from LLM
5xx errors: Retry later

Every error-ku recovery strategy mannumanum!

Q4: Free APIs agents practice-la use panna mudiyuma?

A: Definitely! OpenWeather, NewsAPI, GitHub, JSONPlaceholder, CoinGecko – free tiers available. Perfect for learning! Production: premium APIs select panna-onum, reliability ensure!

Q5: API key security best practice?

Never hardcode .py files-la
Environment variables use (.env)
Production: secrets manager (AWS Secrets, Vault)
Rotate keys regularly
Monitor API usage (detect abuse)
Scope API keys (specific permissions only)

Security = non-negotiable! 🔒

❓ Frequently Asked Questions

❓ API na enna simple la?

Application Programming Interface – oru software innooru software kitta talk panna use pannum method. Restaurant la menu card maari – available options show pannum, neenga order pannunga.

❓ Agent ku API edhuku?

API illama agent just think and talk mattum dhaan pannum. API irundha real actions edukka mudiyum – email send, data fetch, booking, payment, etc.

❓ Free APIs irukka?

Niraiya irukku! OpenWeather, News API, GitHub API, JSONPlaceholder – free tiers available. RapidAPI la 1000+ free APIs browse pannalaam.

❓ API key secure aa store panna epdi?

Environment variables use pannunga (.env file). Never hardcode in source code. Production la use secrets manager (AWS Secrets, HashiCorp Vault).

🧠Knowledge Check

Quiz 1 of 1

Test your API knowledge:

0 of 1 answered

← Previous ByteAgent communication basics Next Byte →Multi-agent architecture

Courses

Learning Paths

Exam Prep

Using APIs inside agents

🔌 Introduction – APIs Give Agents Superpowers!

📚 API Fundamentals for Agents

🔧 How Agents Use APIs – Function Calling

🏗️ API Integration Architecture

🔑 Authentication Methods

🎬 Real Example – Travel Agent APIs

⚠️ Error Handling for APIs

🚀 Popular APIs for Agents

🔄 API Rate Limiting & Optimization

🧪 Try It – Build an API-Using Agent

⚠️ API Security Checklist

💡 API Integration Best Practices

📝 Summary

🏁 🎮 Mini Challenge

💼 Interview Questions

❓ Frequently Asked Questions