Cognee is an open-source AI memory platform. It transforms raw text,
files, and URLs into a persistent knowledge graph that agents can query.

The core workflow has three verbs: remember (ingest data), recall (query
the graph), and forget (evict data). Behind the scenes, remember runs an
extract-cognify-load pipeline: it chunks the input, extracts entities and
relationships with an LLM, embeds the chunks, and stores the results in
a graph database plus a vector store.

For example, Alice works at Anthropic in San Francisco. She contributes
to the Cognee project. The Cognee project depends on litellm for LLM
routing and on Ladybug as its default graph backend. Ladybug is a
lightweight, in-process graph store designed for local development.

Recall queries can traverse this graph: "who works at Anthropic" returns
Alice, "what does Cognee depend on" returns litellm and Ladybug, and
"where does Alice live" returns San Francisco. Recall combines a vector
similarity step with graph traversal so the answers stay grounded in the
original chunks rather than being invented by the language model.
