Skip to main content

​
Recipe Use Cases

This guide covers 12 real-world use cases for PraisonAI recipes, including recommended integration models, implementation steps, and security considerations.

​
1. SaaS AI Support Reply Drafts

​
What It Is

Automatically generate draft replies for customer support tickets using AI, allowing agents to review and send. Model 3 (HTTP Sidecar) or Model 4 (Remote Runner)
  • HTTP API integrates with existing ticketing systems
  • Supports multiple concurrent requests
  • Easy to add auth for multi-tenant SaaS

​
Implementation

1

Create Recipe

# support-reply/TEMPLATE.yaml
schema_version: "1.0"
name: support-reply-drafter
description: Generate support reply drafts

requires:
  env: [OPENAI_API_KEY]

config:
  input:
    ticket_id:
      type: string
      required: true
    customer_message:
      type: string
      required: true
    context:
      type: string
      required: false
2

Start Server

praisonai recipe serve --port 8765 --auth api-key
3

Integrate with Ticketing System

import requests

def generate_reply_draft(ticket_id, customer_message):
    response = requests.post(
        "http://localhost:8765/v1/recipes/run",
        headers={"X-API-Key": "your-key"},
        json={
            "recipe": "support-reply-drafter",
            "input": {
                "ticket_id": ticket_id,
                "customer_message": customer_message
            }
        }
    )
    return response.json()["output"]["draft"]

​
Security Considerations

  • PII: Customer messages may contain PII; ensure data handling compliance
  • Output review: Always have human review before sending
  • Rate limiting: Prevent abuse with request limits

​
Example Payload

{
  "recipe": "support-reply-drafter",
  "input": {
    "ticket_id": "T-12345",
    "customer_message": "I can't log into my account",
    "context": "Premium customer, 2 years"
  }
}

​
2. Meeting Notes → Action Items

​
What It Is

Extract action items, decisions, and follow-ups from meeting transcripts or notes. Model 2 (CLI) or Model 5 (Event-Driven)
  • CLI for ad-hoc processing
  • Event-driven for automated pipeline from recording service

​
Implementation

1

Process via CLI

praisonai recipe run meeting-action-items \
  --input '{"transcript": "...meeting notes..."}' \
  --json
2

Automated Pipeline

# Triggered when new transcript arrives
from praisonai import recipe

def process_meeting(transcript_path):
    with open(transcript_path) as f:
        transcript = f.read()
    
    result = recipe.run(
        "meeting-action-items",
        input={"transcript": transcript}
    )
    
    return result.output["action_items"]

​
Security Considerations

  • Confidentiality: Meeting content may be sensitive
  • Access control: Restrict who can process which meetings
  • Retention: Define data retention policies

​
3. SQL Analyst Assistant

​
What It Is

Natural language to SQL query generation with schema awareness and result explanation. Model 1 (Embedded SDK)
  • Direct integration with data tools
  • Low latency for interactive queries
  • Access to database connections

​
Implementation

1

Define Recipe with Schema

from praisonai import recipe

result = recipe.run(
    "sql-analyst",
    input={
        "question": "Show me top 10 customers by revenue",
        "schema": {
            "customers": ["id", "name", "email"],
            "orders": ["id", "customer_id", "amount", "date"]
        }
    }
)

sql_query = result.output["sql"]
explanation = result.output["explanation"]

​
Security Considerations

  • SQL injection: Validate generated queries before execution
  • Data access: Ensure recipe only sees allowed schemas
  • Audit logging: Log all generated queries

​
4. Code Review Assistant

​
What It Is

Automated code review providing feedback on style, bugs, security, and best practices. Model 6 (Plugin Mode) or Model 2 (CLI)
  • IDE plugin for real-time feedback
  • CLI for CI/CD integration

​
Implementation

1

CI/CD Integration

# In GitHub Actions
- name: AI Code Review
  run: |
    git diff origin/main...HEAD > changes.diff
    praisonai recipe run code-review \
      --input "{\"diff\": \"$(cat changes.diff)\"}" \
      --json > review.json
2

IDE Plugin

// VS Code extension
vscode.commands.registerCommand('praisonai.review', async () => {
  const code = vscode.window.activeTextEditor.document.getText();
  const result = await invokeRecipe('code-review', { code });
  showReviewPanel(result.output.feedback);
});

​
5. Marketing Content Generator

​
What It Is

Generate marketing copy, social posts, email campaigns, and ad variations. Model 3 (HTTP Sidecar)
  • Integrates with marketing platforms
  • Supports batch generation
  • Easy A/B testing

​
Implementation

1

Generate Content

praisonai endpoints invoke marketing-content \
  --input-json '{
    "product": "AI Assistant",
    "tone": "professional",
    "channels": ["email", "linkedin", "twitter"]
  }' \
  --json

​
Security Considerations

  • Brand safety: Review generated content for brand alignment
  • Compliance: Ensure claims are accurate and compliant

​
6. Customer Onboarding Concierge

​
What It Is

Interactive AI assistant guiding new customers through setup, configuration, and first steps. Model 4 (Remote Runner) with streaming
  • Real-time conversational interface
  • Multi-tenant support
  • Session state management

​
Implementation

1

Stream Responses

from praisonai import recipe

for event in recipe.run_stream(
    "onboarding-concierge",
    input={"user_id": "U-123", "step": "initial"},
    session_id="session-abc"
):
    if event.event_type == "progress":
        print(event.data["message"])

​
7. Fraud Signal Summarizer

​
What It Is

Analyze transaction patterns and summarize potential fraud indicators for human review. Model 5 (Event-Driven)
  • Process transactions asynchronously
  • Handle high volume
  • Integrate with alerting systems

​
Implementation

1

Event Handler

def handle_transaction(transaction):
    result = recipe.run(
        "fraud-analyzer",
        input={
            "transaction_id": transaction["id"],
            "amount": transaction["amount"],
            "patterns": transaction["patterns"]
        }
    )
    
    if result.output["risk_score"] > 0.8:
        send_alert(result.output["summary"])

​
Security Considerations

  • Data sensitivity: Transaction data is highly sensitive
  • False positives: Balance detection vs. customer friction
  • Audit trail: Log all decisions for compliance

​
8. Document Ingestion + Q&A

​
What It Is

Ingest documents, build knowledge base, and answer questions with citations. Model 1 (Embedded SDK) or Model 3 (HTTP Sidecar)
  • SDK for direct integration with document pipelines
  • HTTP for multi-service architecture

​
Implementation

1

Ingest Documents

result = recipe.run(
    "doc-ingestion",
    input={"documents": ["/path/to/doc1.pdf", "/path/to/doc2.pdf"]}
)
knowledge_base_id = result.output["kb_id"]
2

Query Knowledge Base

result = recipe.run(
    "doc-qa",
    input={
        "question": "What is the refund policy?",
        "kb_id": knowledge_base_id
    }
)
print(result.output["answer"])
print(result.output["citations"])

​
9. Sales Call Coaching

​
What It Is

Analyze sales call recordings/transcripts and provide coaching feedback. Model 5 (Event-Driven)
  • Process calls asynchronously after completion
  • Integrate with call recording systems
  • Batch processing for historical analysis

​
Implementation

1

Process Call

praisonai recipe run sales-coach \
  --input '{"transcript": "...", "rep_id": "R-123"}' \
  --json

​
10. Data Migration Assistant

​
What It Is

Assist with data transformation, mapping, and validation during migrations. Model 2 (CLI)
  • Scriptable for migration pipelines
  • Batch processing support
  • Easy to integrate with ETL tools

​
Implementation

1

Generate Mapping

praisonai recipe run data-mapper \
  --input '{
    "source_schema": {...},
    "target_schema": {...},
    "sample_data": [...]
  }' \
  --json > mapping.json

​
11. Multi-Agent Router

​
What It Is

Intelligent routing of requests to specialized agents based on intent and context. Model 1 (Embedded SDK) or Model 3 (HTTP Sidecar)
  • Low latency routing decisions
  • Access to multiple downstream agents

​
Implementation

1

Router Recipe

result = recipe.run(
    "agent-router",
    input={
        "user_message": "I need help with billing",
        "available_agents": ["billing", "technical", "sales"]
    }
)

target_agent = result.output["selected_agent"]
# Route to target agent

​
12. Localization Pipeline

​
What It Is

Translate and localize content while preserving context, tone, and formatting. Model 5 (Event-Driven) or Model 2 (CLI)
  • Batch processing for content updates
  • Event-driven for real-time localization

​
Implementation

1

Localize Content

praisonai recipe run localization \
  --input '{
    "content": "Welcome to our platform!",
    "source_lang": "en",
    "target_langs": ["es", "fr", "de", "ja"],
    "context": "marketing_homepage"
  }' \
  --json

​
Security Considerations

  • Cultural sensitivity: Review translations for cultural appropriateness
  • Legal compliance: Ensure translations meet local regulations

​
Testing with Real API Keys

For all use cases, test with real API keys: 
# Set API key
export OPENAI_API_KEY=your-key

# Run a test
praisonai recipe run my-recipe \
  --input '{"test": "data"}' \
  --json

# Verify output
echo $?  # Should be 0

​
Next Steps