Create Custom Recipes

Step-by-step guide to creating your own PraisonAI recipes

Define roles and tasks in agents.yaml, then run the recipe from the CLI.

from praisonaiagents import Agent

agent = Agent(
    name="Recipe Author",
    instructions="Design multi-step recipe roles and validate outputs.",
)
agent.start("Outline a two-agent research recipe.")

The user authors agents.yaml, runs praisonai recipe run, and receives output from each role.

How It Works

---

How to Create a Custom Recipe from Scratch

1. Create Recipe Directory

   Create a new directory for your recipe with the required structure:

   mkdir -p my-recipe
   cd my-recipe
   
2. Create agents.yaml

   Create the main recipe configuration file:

   # agents.yaml
   framework: praisonai
   topic: "{{task}}"
   
   roles:
     researcher:
       role: Research Specialist
       goal: Research the given topic thoroughly
       backstory: Expert researcher with attention to detail
       tools:
         - internet_search
       tasks:
         research_task:
           description: |
             Research: {{task}}
             Provide comprehensive findings.
           expected_output: "Detailed research report"
   
3. Test Your Recipe Locally

   Run your recipe to verify it works:

   praisonai recipe run ./my-recipe --var task="AI trends 2024"
   

How to Create a Recipe with CLI

1. Initialize Recipe

   praisonai recipe init my-new-recipe
   
2. Edit Generated Files

   cd my-new-recipe
   # Edit agents.yaml as needed
   
3. Validate Recipe

   praisonai recipe validate ./my-new-recipe
   
4. Run Recipe

   praisonai recipe run my-new-recipe --var task="Your task here"
   

Recipe File Structure

my-recipe/
├── agents.yaml        # Agent definitions and tasks
├── tools.py           # Optional: custom tools
└── README.md          # Optional: documentation

agents.yaml Schema

Field	Type	Required	Description
framework	string	Yes	Should be praisonai
topic	string	Yes	Main topic/task
roles	object	Yes	Agent definitions
roles.*.role	string	Yes	Agent role name
roles.*.goal	string	Yes	Agent goal
roles.*.tools	array	No	Tools for agent
roles.*.tasks	object	Yes	Agent tasks

Best Practices

Start from praisonai recipe init

The scaffold gives you a valid agents.yaml to edit, so you avoid schema mistakes on the first pass.

Give each role a clear goal and expected_output

Specific goals and expected_output fields keep multi-role recipes on track and make outputs predictable.

Validate before running

Run praisonai recipe validate ./my-recipe after each change so schema errors surface before execution.

---

Related

Add Tools to Recipes

Give recipe agents real capabilities

Different Ways to Create

Other recipe creation methods