Foundations

Context Window vs Memory

A context window is what the model can actively use right now; memory is information that may be stored and retrieved when needed.

Grounded in this concept — answers stay short and honest.

Key terms

Context windowSession memoryPersistent memory
Step 1 of 5

Context window (now)

Active prompt

Recent conversation

Retrieved text

What fits here right now is what the model can use right now.

Memory (stored)

Session state

Saved preferences

Stored summaries

Only useful when the right pieces are retrieved back into the active context.

Business impact

How this shapes cost, speed, risk, and control.

Continuity

Persistent-memory driven

Context alone cannot create cross-session continuity.

Governance impact

High

Persistent memory needs clear policy on what is stored and who can see it.

Cost

Longer context = more tokens

Retrieval lets you keep context tight while reaching broader knowledge.

What can go wrong

Common failure modes to watch for when this concept shows up in production.

Assuming the model remembers everything

Without retrieval, prior conversations or documents are invisible to the model.

Context overflow

Stuffing the window too full crowds out the signal and increases cost.

Wrong memory retrieved

Bad retrieval pulls irrelevant or stale memory into the active prompt.

Related concepts