Cognee gives ChatGPT a memory that lasts beyond one conversation. Everything you ask it to remember goes into your own Cognee Cloud workspace, where it is turned into a knowledge graph: entities, relationships and summaries, not just stored text. Later, in any conversation and from any client connected to the same workspace, ChatGPT can recall it with search that uses that structure, so answers draw on how facts connect rather than on keyword matches alone.
Four tools:
- remember: store text, or a file up to 10 MB, into a dataset. Cognee ingests it, extracts entities and relationships, and builds the graph.
- recall: search your memory with a natural-language question. Cognee routes it to the right retrieval strategy and returns an answer with the supporting memories, or the raw matching passages if you ask for them.
- forget: delete one item, one dataset, or everything you own.
- cognify_status: check on ingestion that was started in the background.
Memories are organised into datasets. By default ChatGPT writes to a dataset of its own, so what ChatGPT remembers stays separate from what your other tools remember, while recall searches across all of them. Name a dataset in the conversation to use a specific one.
You sign in with your Cognee account; no API key is pasted anywhere. Each call runs against your workspace with your own identity and permissions. Workspaces that have been idle go to sleep; the first call after a quiet period wakes one and can take about a minute.
Cognee is open source (github.com/topoteretes/cognee). This plugin is the hosted version of the Cognee MCP server, run by Cognee for Cognee Cloud workspaces. A Cognee Cloud account with a workspace is required.