cognee

We build for developers who need a reliable, production-ready data layer for AI applications

What is cognee?

Cognee implements scalable, modular ECL (Extract, Cognify, Load) pipelines that allow you to interconnect and retrieve past conversations, documents, and audio transcriptions while reducing hallucinations, developer effort, and cost. Try it in a Google Colab notebook or have a look at our documentation

If you have questions, join our Discord community

📦 Installation

With pip

pip install cognee

With pip with PostgreSQL support

pip install cognee[postgres]

With poetry

poetry add cognee

With poetry with PostgreSQL support

poetry add cognee -E postgres

💻 Basic Usage

Setup

import os

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

or

import cognee
cognee.config.set_llm_api_key("YOUR_OPENAI_API_KEY")

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

If you are using Network, create an account on Graphistry to visualize results:

cognee.config.set_graphistry_config({
    "username": "YOUR_USERNAME",
    "password": "YOUR_PASSWORD"
})

(Optional) To run the UI, go to cognee-frontend directory and run:

npm run dev

or run everything in a docker container:

docker-compose up

Then navigate to localhost:3000

If you want to use the UI with PostgreSQL through docker-compose make sure to set the following values in the .env file:

DB_PROVIDER=postgres

DB_HOST=postgres
DB_PORT=5432

DB_NAME=cognee_db
DB_USERNAME=cognee
DB_PASSWORD=cognee

Simple example

Run the default cognee pipeline:

import cognee

text = """Natural language processing (NLP) is an interdisciplinary
       subfield of computer science and information retrieval"""

await cognee.add(text) # Add a new piece of information

await cognee.cognify() # Use LLMs and cognee to create a knowledge graph

search_results = await cognee.search("INSIGHTS", {'query': 'NLP'}) # Query cognee for the insights

for result in search_results:
    do_something_with_result(result)

Create your own memory store

cognee framework consists of tasks that can be grouped into pipelines. Each task can be an independent part of business logic, that can be tied to other tasks to form a pipeline. These tasks persist data into your memory store enabling you to search for relevant context of past conversations, documents, or any other data you have stored.

Example: Classify your documents

Here is an example of how it looks for a default cognify pipeline:

1. To prepare the data for the pipeline run, first we need to add it to our metastore and normalize it:

Start with:

text = """Natural language processing (NLP) is an interdisciplinary
       subfield of computer science and information retrieval"""

await cognee.add(text) # Add a new piece of information

2. In the next step we make a task. The task can be any business logic we need, but the important part is that it should be encapsulated in one function.

Here we show an example of creating a naive LLM classifier that takes a Pydantic model and then stores the data in both the graph and vector stores after analyzing each chunk. We provided just a snippet for reference, but feel free to check out the implementation in our repo.

async def chunk_naive_llm_classifier(
    data_chunks: list[DocumentChunk],
    classification_model: Type[BaseModel]
):
    # Extract classifications asynchronously
    chunk_classifications = await asyncio.gather(
        *(extract_categories(chunk.text, classification_model) for chunk in data_chunks)
    )

    # Collect classification data points using a set to avoid duplicates
    classification_data_points = {
        uuid5(NAMESPACE_OID, cls.label.type)
        for cls in chunk_classifications
    } | {
        uuid5(NAMESPACE_OID, subclass.value)
        for cls in chunk_classifications
        for subclass in cls.label.subclass
    }

    vector_engine = get_vector_engine()
    collection_name = "classification"

    # Define the payload schema
    class Keyword(BaseModel):
        uuid: str
        text: str
        chunk_id: str
        document_id: str

    # Ensure the collection exists and retrieve existing data points
    if not await vector_engine.has_collection(collection_name):
        await vector_engine.create_collection(collection_name, payload_schema=Keyword)
        existing_points_map = {}
    else:
        existing_points_map = {}
    return data_chunks

...

We have many tasks that can be used in your pipelines, and you can also create your tasks to fit your business logic.

3. Once we have our tasks, it is time to group them into a pipeline. This simplified snippet demonstrates how tasks can be added to a pipeline, and how they can pass the information forward from one to another.

            

Task(
    chunk_naive_llm_classifier,
    classification_model = cognee_config.classification_model,
)

pipeline = run_tasks(tasks, documents)

To see the working code, check cognee.api.v1.cognify default pipeline in our repo.

Vector retrieval, Graphs and LLMs

Cognee supports a variety of tools and services for different operations:

• Modular: Cognee is modular by nature, using tasks grouped into pipelines

• Local Setup: By default, LanceDB runs locally with NetworkX and OpenAI.

• Vector Stores: Cognee supports LanceDB, Qdrant, PGVector and Weaviate for vector storage.

• Language Models (LLMs): You can use either Anyscale or Ollama as your LLM provider.

• Graph Stores: In addition to NetworkX, Neo4j is also supported for graph storage.

• User management: Create individual user graphs and manage permissions

Demo

Check out our demo notebook here

Get Started

Install Server

Please see the cognee Quick Start Guide for important configuration information.

docker compose up

Install SDK

Please see the cognee Development Guide for important beta information and usage instructions.

pip install cognee

Star History

💫 Contributors