Decision Guide

When to use which integration model

Pick the integration model that fits your stack, latency budget, and operational constraints.

from praisonaiagents import Agent

agent = Agent(
    name="Integration Advisor",
    instructions="Recommend a PraisonAI recipe integration model for the described app.",
)
agent.start("We need recipes from a Node.js microservice on one machine.")

The user describes their environment; you map it to the right recipe integration pattern below.

Quick Start

1. Simple Usage

   Describe your environment and let the advisor recommend a model.

   from praisonaiagents import Agent
   
   agent = Agent(name="Integration Advisor", instructions="Recommend a recipe integration model.")
   agent.start("We need recipes from a Node.js microservice on one machine.")
   
2. With Configuration

   Once you pick the sidecar model, serve recipes locally.

   praisonai serve recipe --config agents.yaml
   

---

How It Works

---

Quick Decision Tree

Comparison Matrix

Criteria	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
Latency	Lowest	Low	Medium	Medium	Variable	Low
Language	Python	Any	Any	Any	Any	Platform
Complexity	Low	Low	Medium	High	High	Medium
Multi-tenant	❌	❌	❌	✅	✅	❌
Async	❌	❌	❌	❌	✅	❌
Auth built-in	❌	❌	✅	✅	✅	❌
Streaming	✅	✅	✅	✅	❌	✅

By Use Case

Use Case	Primary Model	Alternative
SaaS support replies	Model 3/4	Model 1
Meeting notes	Model 2/5	Model 1
SQL assistant	Model 1	Model 3
Code review	Model 6/2	Model 3
Marketing content	Model 3	Model 5
Customer onboarding	Model 4	Model 3
Fraud detection	Model 5	Model 4
Document Q&A	Model 1/3	Model 4
Sales coaching	Model 5	Model 2
Data migration	Model 2	Model 1
Multi-agent router	Model 1/3	Model 4
Localization	Model 5/2	Model 3

By Environment

Development

• Recommended: Model 1 (SDK) or Model 2 (CLI)
• Why: Fast iteration, easy debugging

Staging

• Recommended: Model 3 (HTTP Sidecar)
• Why: Mimics production, tests HTTP integration

Production

• Recommended: Model 4 (Remote Runner) or Model 5 (Event-Driven)
• Why: Scalable, secure, monitored

Configuration Consolidation

Using agents.yaml for All Config

You can consolidate serve configuration into your existing agents.yaml file instead of maintaining a separate serve.yaml:

# agents.yaml - unified configuration
framework: praisonai
topic: My Application

# Agent definitions
roles:
  assistant:
    role: AI Assistant
    goal: Help users
    tasks:
      respond:
        description: Respond to {query}

# Server configuration (optional section)
serve:
  host: 127.0.0.1
  port: 8765
  auth: api-key
  preload: true
  recipes:
    - my-recipe
    - another-recipe

Then start the server:

praisonai serve recipe --config agents.yaml

This approach:

• ✅ Single configuration file
• ✅ Agent definitions + server config together
• ✅ Easier to manage
• ✅ Version control friendly

Best Practices

Let language decide the first branch

Python apps get the Embedded SDK’s lowest latency; anything else starts at CLI, Sidecar, or Remote Runner.

Match the model to the environment, not just the app

Dev favours SDK/CLI for fast iteration; production favours Remote Runner or Event-Driven for scale, auth, and monitoring.

Consolidate config in agents.yaml

A serve: section in agents.yaml keeps agent and server settings in one version-controlled file instead of a separate serve.yaml.

Reach for async models for batch work

Event-Driven is the only model with built-in async, so choose it when throughput and queued processing matter more than latency.

---

Related

Integration Models

Detailed setup for each model

Use Cases

See which model fits each scenario