Hello cognee SDK

A quick tutorial to get started with cognee SDK using different configurations.

Check our .env.template for all configuration options.

Install cognee SDK


$ uv venv
$ source .venv/bin/activate
$ uv pip install cognee

Add your OpenAI key to cognee

Create a .env file in your project directory:


echo 'LLM_API_KEY="your_api_key"' > .env

Configure cognee via ENV variables

Cognee supports multiple LLM providers and database configurations. Here are the main options:

LLM Providers

OpenAI: Simple setup with LLM_API_KEY
Azure OpenAI: Use LLM_MODEL="azure/gpt-4o-mini" with endpoint and API version
Ollama: Free local LLMs with LLM_PROVIDER="ollama"
OpenRouter: Free models with LLM_PROVIDER="custom"
DeepInfra: Custom provider setup

Database Options

Default: SQLite + NetworkX + LanceDB (no setup required)
Neo4j: For advanced graph operations
PostgreSQL: For production relational storage
PGVector: For PostgreSQL-based vector storage

Case 0 - Default cognee

The simplest setup using local storage (no external services required):


# .env
LLM_API_KEY="your_openai_key"
 
# Optional: Specify default providers explicitly
GRAPH_DATABASE_PROVIDER="networkx"
VECTOR_DB_PROVIDER="lancedb"
DB_PROVIDER="sqlite"
DB_NAME="cognee_db"

Cognee stores files in .cognee/ directory by default. Use DATA_ROOT_DIRECTORY and SYSTEM_ROOT_DIRECTORY to customize storage locations.


import cognee
import asyncio
 
async def main():
    text = "Cognee is a memory layer for AI agents."
    await cognee.add(text)
    await cognee.cognify()
    
    results = await cognee.search("What is cognee?")
    print(results[0])
 
asyncio.run(main())

Case 1 - Using Neo4j

For advanced graph capabilities with Neo4j:


# .env
LLM_API_KEY="your_openai_key"
 
# Neo4j configuration
GRAPH_DATABASE_PROVIDER="neo4j"
GRAPH_DATABASE_URL="bolt://localhost:7687"
GRAPH_DATABASE_USERNAME="neo4j"
GRAPH_DATABASE_PASSWORD="your_password"
 
# Keep other defaults
VECTOR_DB_PROVIDER="lancedb"
DB_PROVIDER="sqlite"

Start Neo4j locally:


docker run -p 7474:7474 -p 7687:7687 -e NEO4J_AUTH=neo4j/your_password neo4j:latest

Case 2 - Using PostgreSQL + PGVector

For production-ready setup with PostgreSQL:


# .env
LLM_API_KEY="your_openai_key"
 
# PostgreSQL configuration
DB_PROVIDER="postgres"
DB_NAME="cognee_db"
DB_HOST="127.0.0.1"
DB_PORT="5432"
DB_USERNAME="cognee"
DB_PASSWORD="cognee"
 
# PGVector for embeddings
VECTOR_DB_PROVIDER="pgvector"
 
# Graph database
GRAPH_DATABASE_PROVIDER="kuzu"

Free Local Setup

Run cognee completely offline with Ollama:


# .env
LLM_API_KEY="ollama"
LLM_MODEL="llama3.1:8b"
LLM_PROVIDER="ollama"
LLM_ENDPOINT="http://localhost:11434/v1"
 
EMBEDDING_PROVIDER="ollama"
EMBEDDING_MODEL="avr/sfr-embedding-mistral:latest"
EMBEDDING_ENDPOINT="http://localhost:11434/api/embeddings"
EMBEDDING_DIMENSIONS="4096"

Install and start Ollama:


# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh
 
# Pull models
ollama pull llama3.1:8b
ollama pull avr/sfr-embedding-mistral:latest

Next Steps

Explore custom pipelines
Learn about data ingestion
Check out search types