Memory, context & research

topoteretes/cognee

Persistent, self-hostable graph memory for agents

30k
Apache-2.0
Python

What it is and why it matters

Open-source memory platform for agents that replaces flat vector search with a dynamic knowledge graph: it retrieves context and also models the relationships between data. It connects to sources like Notion or Slack and keeps memory updated across sessions. Per its authors, it resolves conflicts by retracing which decisions came later rather than piling up unstructured context.

Key points

Dynamic memory graph versus flat vector memory
Connects to company data via agents, Notion and Slack
Self-hostable knowledge graph engine, Apache-2.0 licensed

Where we saw it

cognee 2026-07-17 The launch video contrasts a Claude Code session with and without Cognee: without structured memory the agent misses a recent client change; with Cognee it recognizes the most up-to-date fact and corrects its own memory. The accuracy and cost figures are the project's own claims. cognee 1.0: Self-improving memory for agents

Same category

Graphify-Labs/graphify Navigable code knowledge graph, no vector store Open detail headroomlabs-ai/headroom Compress tool outputs before they reach the LLM Open detail mvanhorn/last30days-skill Research the last 30 days across X, Reddit and more Open detail