topoteretes/cognee is a self-hostable AI memory platform that ingests data in any format to build a knowledge graph for persistent long-term memory across sessions. A managed hosted service is also available, though operation requires an external LLM API provider for inference.
Project overview
The platform enables running the entire memory layer, including graph relationships, embeddings, sessions, and metadata, on a single Postgres instance.
Project type
AI Agent · Vector Database
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Developers and enterprise teams who need a self-hosted knowledge graph to provide persistent long-term memory for AI agents.
AI engineers who want to build retrieval augmented generation pipelines using graph relationships, vector embeddings, and ontology generation.
Key capabilities
Gives AI agents persistent long-term memory across sessions by continuously building a self-hosted knowledge graph.
Combines vector embeddings, graph reasoning, and cognitive-science-grounded ontology generation to make documents searchable by meaning and connected by relationships.
Ingests data in any format to unify data from various sources in one place.
Provides four API operations to manage agent memory: remember, recall, forget, and improve.
Allows running the entire memory layer including graph relationships, embeddings, sessions, and metadata on a single Postgres instance.
Limitations and risks
Requires an external LLM API key to function.
Depends on external Large Language Models for processing data and managing memory.
Getting started
Install the cognee library via pip. Configure an LLM_API_KEY, then use the cognee-cli to remember or recall information.
Alternatives and comparisons
Builds and queries temporal context graphs that track how facts change over time. Relevant if you need incremental updates without complete graph recomputation.
Provides 100+ open-source AI agent and RAG app examples to learn from and build upon.