Skip to main content
A multi-modal RAG agent that retrieves recipes from a knowledge base and generates step-by-step visual instruction manuals using image generation.

​
Key Concepts

This agent combines two powerful techniques:
  • RAG (Retrieval Augmented Generation): Searches a vector database for recipes
  • Image Generation: Creates visual guides using DALL-E via OpenAITools

​
Prerequisites

  • Python 3.12+
  • Docker (for PostgreSQL with pgvector)
  • OpenAI API key (for GPT and DALL-E)
  • Cohere API key (for embeddings)

​
Setup

1

Clone the repository

git clone https://github.com/agno-agi/agno.git
cd agno
2

Create and activate virtual environment

uv venv --python 3.12
source .venv/bin/activate
3

Install dependencies

uv pip install -r cookbook/01_showcase/01_agents/recipe_agent/requirements.in
4

Set environment variables

export OPENAI_API_KEY=sk-***
export COHERE_API_KEY=***
5

Start PostgreSQL with pgvector

./cookbook/scripts/run_pgvector.sh
6

Load recipes into knowledge base

python cookbook/01_showcase/01_agents/recipe_agent/scripts/load_recipes.py

​
Run the Agent

​
Basic Recipe

Retrieve a recipe from the knowledge base: 
python cookbook/01_showcase/01_agents/recipe_agent/examples/basic_recipe.py
Demonstrates:
  • Searching the recipe knowledge base
  • Formatted recipe output
  • Ingredient and instruction extraction

​
Visual Guide

Generate a recipe with step-by-step images: 
python cookbook/01_showcase/01_agents/recipe_agent/examples/visual_guide.py
Demonstrates:
  • DALL-E image generation for cooking steps
  • Saving images to disk
  • Complete visual recipe guide

​
Agent Configuration

recipe_agent = Agent(
    name="Recipe Agent",
    model=OpenAIResponses(id="gpt-5.2"),
    system_message=SYSTEM_MESSAGE,
    knowledge=recipe_knowledge,
    tools=[
        OpenAITools(),
        ReasoningTools(add_instructions=True),
    ],
    add_datetime_to_context=True,
    add_history_to_context=True,
    num_history_runs=5,
    enable_agentic_memory=True,
    search_knowledge=True,
    markdown=True,
)
ParameterPurpose
modelGPT-5.2 for recipe understanding and generation
knowledgeVector database with recipe documents
OpenAIToolsDALL-E image generation
search_knowledgeEnable knowledge base search
ReasoningToolsPlan recipe presentation

​
How It Works

​
Recipe Workflow

1. User requests a recipe
2. Agent searches knowledge base for matches
3. Retrieves and formats recipe details
4. Identifies 2-3 key visual steps
5. Generates images for each step using DALL-E
6. Returns combined recipe text and images

​
Knowledge Base

Recipes are stored in PostgreSQL with pgvector using Cohere embeddings: 
recipe_knowledge = Knowledge(
    name="Recipe Knowledge Base",
    vector_db=PgVector(
        db_url=DB_URL,
        table_name="recipe_documents",
        embedder=CohereEmbedder(id="embed-v4.0"),
    ),
    max_results=5,
)

​
Image Generation

The agent generates images for key cooking steps:
Step TypeExample Prompt
Ingredient prep”Thai green curry ingredients on cutting board, food photography”
Cooking technique”Wok with stir-fried vegetables, steam rising, overhead shot”
Final dish”Plated Thai curry with jasmine rice, garnished with basil”

​
Troubleshooting

Ensure recipes are loaded into the knowledge base:
python cookbook/01_showcase/01_agents/recipe_agent/scripts/load_recipes.py
Check your OpenAI API key has DALL-E access. Image generation requires a paid OpenAI account.
Ensure PostgreSQL is running:
docker ps | grep pgvector
./cookbook/scripts/run_pgvector.sh

​
Source Code