cognee - memory layer for AI apps and Agents

Demo . Learn more · Join Discord

AI Agent responses you can rely on.

Build dynamic Agent memory using scalable, modular ECL (Extract, Cognify, Load) pipelines.

More on use-cases.

Features

• Interconnect and retrieve your past conversations, documents, images and audio transcriptions
• Reduce hallucinations, developer effort, and cost.
• Load data to graph and vector databases using only Pydantic
• Manipulate your data while ingesting from 30+ data sources

Get Started

Get started quickly with a Google Colab notebook or starter repo

Contributing

Your contributions are at the core of making this a true open source project. Any contributions you make are greatly appreciated. See CONTRIBUTING.md for more information.

📦 Installation

You can install Cognee using either pip, poetry, uv or any other python package manager.

With pip

pip install cognee

💻 Basic Usage

Setup

import os
os.environ["LLM_API_KEY"] = "YOUR OPENAI_API_KEY"

You can also set the variables by creating .env file, using our template. To use different LLM providers, for more info check out our documentation

Simple example

Add LLM_API_KEY to .env using the command bellow.

echo "LLM_API_KEY=YOUR_OPENAI_API_KEY" > .env

You can see available env variables in the repository .env.template file. If you don't specify it otherwise, like in this example, SQLite (relational database), LanceDB (vector database) and NetworkX (graph store) will be used as default components.

This script will run the default pipeline:

import cognee
import asyncio


async def main():
    # Add text to cognee
    await cognee.add("Natural language processing (NLP) is an interdisciplinary subfield of computer science and information retrieval.")

    # Generate the knowledge graph
    await cognee.cognify()

    # Query the knowledge graph
    results = await cognee.search("Tell me about NLP")

    # Display the results
    for result in results:
        print(result)


if __name__ == '__main__':
    asyncio.run(main())

Example output:

       # ({'id': UUID('bc338a39-64d6-549a-acec-da60846dd90d'), 'updated_at': datetime.datetime(2024, 11, 21, 12, 23, 1, 211808, tzinfo=datetime.timezone.utc), 'name': 'natural language processing', 'description': 'An interdisciplinary subfield of computer science and information retrieval.'}, {'relationship_name': 'is_a_subfield_of', 'source_node_id': UUID('bc338a39-64d6-549a-acec-da60846dd90d'), 'target_node_id': UUID('6218dbab-eb6a-5759-a864-b3419755ffe0'), 'updated_at': datetime.datetime(2024, 11, 21, 12, 23, 15, 473137, tzinfo=datetime.timezone.utc)}, {'id': UUID('6218dbab-eb6a-5759-a864-b3419755ffe0'), 'updated_at': datetime.datetime(2024, 11, 21, 12, 23, 1, 211808, tzinfo=datetime.timezone.utc), 'name': 'computer science', 'description': 'The study of computation and information processing.'})

For more advanced usage, have a look at our documentation.

Understand our architecture

Demos

What is AI memory:

Learn about cognee

Code of Conduct

We are committed to making open source an enjoyable and respectful experience for our community. See CODE_OF_CONDUCT for more information.

💫 Contributors

Star History