by Topoteretes

Cognee

Name: Cognee
Price: 35 USD
Author: Topoteretes

Open-source knowledge engine that gives AI agents persistent, self-improving memory using a hybrid graph and vector architecture.

Cognee is an open-source AI memory engine that replaces custom knowledge graphs and vector stores with a single platform. Its ECL pipeline (Extract, Cognify, Load) ingests data from 38+ sources — documents, PDFs, Slack threads, audio, images — and structures it into a queryable knowledge graph with embeddings and relationships. Combines three storage layers: relational, vector, and graph. Achieves up to 92.5% response accuracy via hybrid graph+vector retrieval. Pricing: Free (28+ data sources, community support), Developer ($35/mo, 1,000 docs, hosted on AWS/GCP/Azure), Cloud Team ($200/mo, 2,500 docs, 10 users), Enterprise (custom, on-prem). 16,100 GitHub stars. $7.5M seed raised 2026. Used in production at 70+ companies.

Capabilities

deep-researchdata-analysisopen-sourcebyokautonomouscitations

Pros & Limitations

Editorial assessment

Pros

✓Hybrid graph + vector memory at every pricing tier
✓38+ data connectors including documents, audio, images, Slack
✓Up to 92.5% retrieval accuracy vs ~60% for basic RAG
✓Self-improving memory — learns from feedback over time
✓$7.5M seed backed, 16,100 GitHub stars

Limitations

⚠Better for institutional knowledge than conversation personalization
⚠No published LongMemEval benchmark score for direct comparison
⚠Complex multi-layer architecture has steeper learning curve than simpler memory tools