Basic memory with Cognee

This example demonstrates how to set up Cognee to build a persistent memory layer using a knowledge engine approach (Vector search + Knowledge Graph).

A note on the Cognee API

This example uses add(), cognify(), search(), and the prune helpers, which is the API surface described in Chapter 5. Recent Cognee releases also expose a higher-level surface built around remember(), recall(), forget(), and improve(), which wraps the same ingestion and retrieval pipeline behind fewer calls. Both work with the pinned dependency. The lower-level calls are kept here because they make each stage of the memory pipeline visible, which is the point of the example.

Requirements

• Python 3.10+
• A model API key exposed as LLM_API_KEY (for example, an OpenAI API key)

Steps for running this example in the shell

1. Install dependencies:

python -m venv .venv

# macOS/Linux:
source .venv/bin/activate

# Windows Command Prompt:
.venv\Scripts\activate.bat

# Windows PowerShell:
.venv\Scripts\Activate.ps1

pip install -r requirements.txt

2. Export your model API key as an environment variable:

export LLM_API_KEY="sk-..." # Windows cmd: set LLM_API_KEY="sk-..." # Windows PowerShell: $env:LLM_API_KEY="sk-..."

3. Run the script:

python cognee_memory.py

Output

When you run the script, it will add a short text piece to the Cognee memory layer, "cognify" it (processing it into a knowledge graph and vectorizing it), and then perform a hybrid search to retrieve it.

=== Cognee Memory Example ===
Pruning old data...
Storage manager absolute path: ...\.venv\Lib\site-packages\cognee\.cognee_cache
Deleting cache...
✓ Cache deleted successfully!
Adding context to memory...
User 48a79c34-ee78-4f81-afcc-81947dc8a4df has registered.
Data added: 'Context engineering is the process of optimizing information provided to an LLM to improve its performance and accuracy.'
Cognifying (building knowledge graph and vector index)...
Searching memory for: 'What is context engineering?'
--- Memory Retrieval Results ---
Result 1: {'dataset_id': UUID('db5e4b38-8ece-5953-bbf1-dd723eee7283'), 'dataset_name': 'main_dataset', 'dataset_tenant_id': None, 'search_result': ['Context engineering is the process of optimizing the information supplied to a large language model (LLM) to improve its effectiveness, accuracy, and overall performance.']}