# Memory vs RAG

Cortex is often mistaken for a RAG (retrieval-augmented generation) stack because both
"fetch relevant context." They solve different problems. RAG retrieves chunks; Cortex
**remembers**.

## The core difference

A RAG pipeline indexes documents and returns the closest chunks to a query. It is
stateless about *you*: it has no notion of what is true now, what was corrected, or what
should be forgotten. Cortex maintains **durable, evolving state**: structured memories
with provenance, confidence, and a version history that consolidation keeps clean.

| | RAG pipeline | **Cortex memory** |
| --- | --- | --- |
| Unit | Document chunk | Structured [`Memory`](/concepts/memory-model) |
| State | Stateless index | Durable, versioned namespace |
| Truth | All chunks equal | Verified-first, confidence-scored |
| Corrections | New doc competes with old | Consolidation makes the update current truth |
| Forgetting | Re-index everything | Tombstone; recall stops surfacing it |
| Evolution | Static until re-indexed | [Elastic brain](/concepts/elastic-brain) folds & fades over time |
| Ownership | Wherever the vector DB lives | Sovereign, encrypted, user-controlled |

## What this means in practice

:::note\[Corrections, not contradictions]
In a vector store, an updated fact just becomes another nearby chunk, and the old one is
still there, still retrievable, still competing. In Cortex, consolidation resolves the
two so the correction becomes the **current** fact.
:::

* **Recall is verified-first.** Trusted memories surface ahead of unconfirmed ones, rather than ranking purely on similarity.
* **Memory has provenance.** Every memory traces back to its source, so you can see *why* something is known.
* **Memory decays gracefully.** Weak, unused signals fade; durable ones get promoted, so your graph stays useful instead of bloating.

## They're not mutually exclusive

Cortex still does retrieval; recall is how memories reach a prompt. The difference is
*what gets retrieved*: not arbitrary chunks, but durable, consolidated, owned memory. You
can think of Cortex as RAG with a memory model underneath it.

## Keep reading

* The artifacts that make this possible: [The Memory Model](/concepts/memory-model).
* How memory self-maintains: [The Elastic Brain](/concepts/elastic-brain).