Praison AI

Agent Recipes Integration

Integrate Agent Recipes into your Python applications

Agent Recipes Integration

Learn how to use Agent Recipes programmatically in your Python applications.

Installation

bash
pip install praisonai agent-recipes praisonai-tools

Basic Usage

Using the LLM Tool Directly

python
import os
os.environ['OPENAI_API_KEY'] = 'your-api-key'

from praisonai_tools.recipe_tools import LLMTool

# Initialize LLM tool
llm = LLMTool(provider="openai", model="gpt-4o-mini")

# Simple completion
response = llm.complete("Write a haiku about coding")
print(response.content)
print(f"Tokens used: {response.usage['total_tokens']}")

Using Multiple Providers

python
from praisonai_tools.recipe_tools import LLMTool

# OpenAI
openai_llm = LLMTool(provider="openai", model="gpt-4o-mini")

# Anthropic
anthropic_llm = LLMTool(provider="anthropic", model="claude-3-haiku-20240307")

# Google
google_llm = LLMTool(provider="google", model="gemini-1.5-flash")

# Use any provider
response = openai_llm.complete("Hello!")

JSON Extraction

python
from praisonai_tools.recipe_tools import LLMTool

llm = LLMTool(provider="openai")

# Extract structured data
schema = {"name": "string", "age": "number", "skills": "array"}
result = llm.extract_json(
    "John is 30 years old and knows Python, JavaScript, and Go",
    schema=schema
)
print(result)
# {'name': 'John', 'age': 30, 'skills': ['Python', 'JavaScript', 'Go']}

Using Recipe Tools

Vision Tool

python
from praisonai_tools.recipe_tools import VisionTool

vision = VisionTool(provider="openai")

# Analyze an image
result = vision.analyze("photo.jpg", prompt="Describe this image")
print(result.description)

# Generate caption
caption = vision.caption("photo.jpg")
print(caption)

# Extract tags
tags = vision.tag("photo.jpg")
print(tags)  # ['nature', 'landscape', 'mountains']

Chart Tool

python
from praisonai_tools.recipe_tools import ChartTool

chart = ChartTool()

# Create bar chart from data
data = {"Q1": 100, "Q2": 150, "Q3": 200, "Q4": 180}
result = chart.bar(data, title="Quarterly Sales", output="sales.png")
print(f"Chart saved to: {result.path}")

# Create from CSV
result = chart.line("data.csv", x="date", y="value", output="trend.png")

Email Tool

python
from praisonai_tools.recipe_tools import EmailTool

email = EmailTool()

# Parse email file
parsed = email.parse("message.eml")
print(f"From: {parsed.sender}")
print(f"Subject: {parsed.subject}")
print(f"Attachments: {len(parsed.attachments)}")

# Extract structured data with LLM
extracted = email.extract("message.eml", fields=["action_items", "deadlines"])
print(extracted)

Loading and Running Recipes

Load Recipe Configuration

python
import yaml
from pathlib import Path

def load_recipe(name: str):
    """Load a recipe's TEMPLATE.yaml configuration."""
    recipe_path = Path.home() / ".praison" / "templates" / name
    if not recipe_path.exists():
        recipe_path = Path("/path/to/your/Agent-Recipes/agent_recipes/templates") / name

    template_file = recipe_path / "TEMPLATE.yaml"
    with open(template_file) as f:
        return yaml.safe_load(f)

# Load recipe
recipe = load_recipe("ai-blog-generator")
print(f"Recipe: {recipe['name']}")
print(f"Description: {recipe['description']}")
print(f"Required tools: {recipe['requires']['tools']}")

Check Dependencies

python
import os
import shutil

def check_dependencies(recipe: dict) -> dict:
    """Check if recipe dependencies are satisfied."""
    results = {}
    requires = recipe.get("requires", {})

    # Check environment variables
    for env_var in requires.get("env", []):
        results[f"env:{env_var}"] = bool(os.environ.get(env_var))

    # Check Python packages
    for package in requires.get("packages", []):
        try:
            __import__(package.replace("-", "_").split("[")[0])
            results[f"package:{package}"] = True
        except ImportError:
            results[f"package:{package}"] = False

    # Check external tools
    for tool in requires.get("external", []):
        results[f"external:{tool}"] = shutil.which(tool) is not None

    return results

# Check dependencies
deps = check_dependencies(recipe)
for dep, status in deps.items():
    print(f"{'✓' if status else '✗'} {dep}")

Building Custom Recipes

Create a Custom Recipe

python
from praisonai_tools.recipe_tools import LLMTool

class BlogGenerator:
    """Custom blog post generator recipe."""

    def __init__(self, provider="openai", model="gpt-4o-mini"):
        self.llm = LLMTool(provider=provider, model=model)

    def generate(self, topic: str, style: str = "professional") -> str:
        """Generate a blog post on the given topic."""
        prompt = f"""Write a blog post about: {topic}

Style: {style}
Include:
- Engaging introduction
- 3-4 main sections with headers
- Practical examples
- Call to action

Format as Markdown."""

        response = self.llm.complete(
            prompt,
            system="You are an expert blog writer.",
            max_tokens=2000
        )
        return response.content

# Use custom recipe
generator = BlogGenerator()
post = generator.generate("AI in Healthcare", style="conversational")
print(post)

Create a Multi-Tool Recipe

python
from praisonai_tools.recipe_tools import LLMTool, ChartTool

class ReportGenerator:
    """Generate reports with charts from data."""

    def __init__(self):
        self.llm = LLMTool(provider="openai")
        self.chart = ChartTool()

    def generate(self, data_file: str, output_dir: str = "./report"):
        import pandas as pd
        import os

        os.makedirs(output_dir, exist_ok=True)

        # Load data
        df = pd.read_csv(data_file)

        # Generate summary with LLM
        summary_prompt = f"""Analyze this data and provide key insights:

Columns: {list(df.columns)}
Sample: {df.head().to_string()}
Stats: {df.describe().to_string()}"""

        summary = self.llm.complete(summary_prompt)

        # Generate chart
        numeric_cols = df.select_dtypes(include=['number']).columns
        if len(numeric_cols) > 0:
            self.chart.bar(
                df[numeric_cols[0]].to_dict(),
                title=f"Distribution of {numeric_cols[0]}",
                output=f"{output_dir}/chart.png"
            )

        # Write report
        with open(f"{output_dir}/report.md", "w") as f:
            f.write(f"# Data Report\n\n{summary.content}\n")

        return {"summary": summary.content, "chart": f"{output_dir}/chart.png"}

# Use multi-tool recipe
reporter = ReportGenerator()
result = reporter.generate("sales.csv")

Error Handling

python
from praisonai_tools.recipe_tools import LLMTool
from praisonai_tools.recipe_tools.base import DependencyError

try:
    llm = LLMTool(provider="openai")
    deps = llm.check_dependencies()

    if not deps.get("api_key"):
        raise DependencyError("OPENAI_API_KEY not set")

    response = llm.complete("Hello")

except DependencyError as e:
    print(f"Missing dependency: {e}")
    print("Install with: pip install openai")
    print("Set: export OPENAI_API_KEY=your-key")

except Exception as e:
    print(f"Error: {e}")

Integration with PraisonAI Agents

python
from praisonaiagents import Agent
from praisonai_tools.recipe_tools import LLMTool

# Create agent with recipe tool
llm_tool = LLMTool(provider="openai")

agent = Agent(
    name="Content Creator",
    instructions="You create engaging content using AI tools.",
    tools=[llm_tool.complete]
)

# Run agent
result = agent.run("Create a social media post about our new product launch")
print(result)

Best Practices

  1. Always check dependencies before running recipes
  2. Use environment variables for API keys
  3. Handle errors gracefully with try/except
  4. Use dry-run mode for testing
  5. Cache results when appropriate
  6. Log execution metadata for debugging

Next Steps