AI Memory

Memory is how an AI product carries selected information into future conversations.

A fresh conversation does not automatically contain everything from earlier sessions. Continuity appears when a product saves or retrieves information and places it back into the current context.

Watch information carry forward

Earlier conversations

Priya is vegetarian, prefers concise answers and is working on the Q3 review.

💾
Today's context

Nothing carried forward

The new conversation only knows what is included now.

Context

The information available to the model right now.

Memory

The product mechanism that decides what can persist and what should be loaded later.

Key insight: memory is not a hidden copy of every past chat. It is a product mechanism for preserving or retrieving selected information.
AI Memory · Go deeper

Products can create continuity in several different ways.

Saved facts, summaries and retrieval solve different memory problems. They all work by bringing selected information back into context.

Four continuity patterns

1

No saved memory

Only the current context is available.

2

Saved facts

Stable preferences or profile details are reloaded.

3

Conversation summary

Key decisions and open items survive without copying every old turn.

4

Smart lookup

Relevant history or documents are retrieved only when needed.

Where memory can fail

1
Nothing was savedThe product cannot retrieve information it never preserved.
2
The wrong thing was savedA stale or incorrect preference can keep influencing later responses.
3
Retrieval missesThe information exists but is not selected for this turn.

What good memory looks like

📝
Save durable facts carefullyRole, preferred format and stable preferences.
🗜
Summarise long historiesPreserve decisions and open items.
🔍
Retrieve only what is relevantAvoid loading every past detail.
✓
Let users review and correctEspecially when memory affects important work.
Next foundation: API. Continue →