Duration	1.5 hours
Day	5 of 5

Learning Objectives

By the end of this module, students will be able to:

• Deploy agents to AWS Lambda
• Configure Google Cloud Functions
• Set up Azure Functions
• Handle cold starts appropriately

Topics

1. Serverless Benefits (15 min)

Why Serverless?

Benefit	Description
No server management	Cloud handles infrastructure
Auto-scaling	Scales with demand
Pay per use	Only pay when called
High availability	Built-in redundancy

When to Use Serverless

Good For	Not Good For
Sporadic traffic	Consistent high load
Variable load	Very low latency requirements
Simple deployments	Complex state management
Cost optimization	Long-running processes

---

2. AWS Lambda (30 min)

Handler Code

# handler.py
from signalwire_agents import AgentBase, SwaigFunctionResult

class MyAgent(AgentBase):
    def __init__(self):
        super().__init__(name="lambda-agent")
        self.prompt_add_section("Role", "You are a helpful assistant.")
        self.add_language("English", "en-US", "rime.spore")

    @AgentBase.tool(description="Get help")
    def get_help(self, args: dict, raw_data: dict = None) -> SwaigFunctionResult:
        return SwaigFunctionResult("How can I help you?")

# Create agent instance
agent = MyAgent()

# Lambda handler
def lambda_handler(event, context):
    return agent.handle_lambda(event, context)

serverless.yml (Serverless Framework)

service: voice-agent

provider:
  name: aws
  runtime: python3.11
  region: us-east-1
  environment:
    SIGNALWIRE_SPACE_NAME: ${env:SIGNALWIRE_SPACE_NAME}
    SIGNALWIRE_PROJECT_ID: ${env:SIGNALWIRE_PROJECT_ID}
    SIGNALWIRE_TOKEN: ${env:SIGNALWIRE_TOKEN}

functions:
  agent:
    handler: handler.lambda_handler
    events:
      - http:
          path: /agent
          method: post
      - http:
          path: /agent/swaig
          method: post
    timeout: 30
    memorySize: 512

plugins:
  - serverless-python-requirements

custom:
  pythonRequirements:
    dockerizePip: true

Deploying

# Install dependencies
npm install -g serverless
npm install serverless-python-requirements

# Deploy
serverless deploy

# Get URL
serverless info

---

3. Google Cloud Functions (25 min)

Handler Code

# main.py
from signalwire_agents import AgentBase, SwaigFunctionResult

class MyAgent(AgentBase):
    def __init__(self):
        super().__init__(name="gcf-agent")
        self.prompt_add_section("Role", "You are a helpful assistant.")
        self.add_language("English", "en-US", "rime.spore")

agent = MyAgent()

def handle_request(request):
    """Google Cloud Function entry point."""
    return agent.handle_cloud_function(request)

requirements.txt

signalwire-agents>=1.0.7

Deploying

gcloud functions deploy voice-agent \
  --runtime python311 \
  --trigger-http \
  --allow-unauthenticated \
  --entry-point handle_request \
  --set-env-vars SIGNALWIRE_SPACE_NAME=your-space,SIGNALWIRE_PROJECT_ID=your-id,SIGNALWIRE_TOKEN=your-token

---

4. Azure Functions (20 min)

Handler Code

# __init__.py
import azure.functions as func
from signalwire_agents import AgentBase, SwaigFunctionResult

class MyAgent(AgentBase):
    def __init__(self):
        super().__init__(name="azure-agent")
        self.prompt_add_section("Role", "You are a helpful assistant.")
        self.add_language("English", "en-US", "rime.spore")

agent = MyAgent()

def main(req: func.HttpRequest) -> func.HttpResponse:
    return agent.handle_azure_function(req)

function.json

{
  "scriptFile": "__init__.py",
  "bindings": [
    {
      "authLevel": "anonymous",
      "type": "httpTrigger",
      "direction": "in",
      "name": "req",
      "methods": ["post"]
    },
    {
      "type": "http",
      "direction": "out",
      "name": "$return"
    }
  ]
}

---

5. Cold Start Optimization (10 min)

What is Cold Start?

First request after idle period:

1. Container starts
2. Runtime loads
3. Dependencies import
4. Code initializes

Can add 1-5 seconds!

Mitigation Strategies

1. Keep Warm

# serverless.yml
functions:
  agent:
    events:
      - schedule: rate(5 minutes)  # Keep warm

2. Minimize Dependencies

# Import only what you need
from signalwire_agents import AgentBase  # Good
# from signalwire_agents import *  # Bad - loads everything

3. Lazy Loading

class LazyAgent(AgentBase):
    _db_client = None

    @property
    def db(self):
        if self._db_client is None:
            from database import Client
            self._db_client = Client()
        return self._db_client

4. Provisioned Concurrency (AWS)

functions:
  agent:
    provisionedConcurrency: 2  # Keep 2 instances warm

---

Key Takeaways

1. Serverless = no server management - Cloud handles it
2. Each platform has handlers - Lambda, GCF, Azure
3. Cold starts are real - Plan for them
4. Keep dependencies minimal - Faster startup
5. Use for variable load - Pay for what you use

Preparation for Lab 2.11

• AWS, GCP, or Azure account
• CLI tools installed
• Environment variables ready

Lab Preview

In Lab 2.11, you will:

1. Deploy to AWS Lambda
2. Test cold start performance
3. Configure environment
4. Connect to SignalWire