What is AI memory?
AI memory is what stays with a tool between one chat and the next. Learn the types, how it works in ChatGPT, Claude and Gemini, and when you need a separate memory.
Guide
AI memory vs context window: what is the difference?
Three terms that get mixed up: context window, memory, and retrieval from documents (RAG). This guide separates them.
Last reviewed:
The short answer
The context window is how much text a model can read in one chat. It is temporary and ends with the chat. Memory is information stored outside the chat that comes back in later chats. A bigger window is not a memory.
The difference in one table
| Context window | Memory | RAG | |
|---|---|---|---|
| What it is | What the model reads right now | What is kept about you and your work | A search over files whose results are added to the question |
| How long it lasts | Until the chat ends | Across chats, until you edit it | As long as the files stay |
| Who writes it | You and the model during the chat | You and the tool over time | You, by uploading files |
| Does it change? | Wiped with each new chat | Builds up and gets updated | Fixed until you replace a file |
Why is a large context window not enough?
Context windows grow with each model generation, and some now hold whole books. But they start empty in every new chat. You fill them again each time.
Even within one chat, the longer it runs, the less attention the model pays to what came first.
And RAG?
RAG means you upload files and the tool searches them on each question, adding what it finds to the context. It suits fixed references: a manual, a contract, a report.
Memory differs because it changes with your work: new decisions are added, old ones corrected, and the model itself writes to it.
How the three fit together
On each request, the tool searches your memory and files, puts what matters into the context window, and the model answers. Memory is the store; the window is the desk the work happens on.
Common questions
How big is the context window in ChatGPT, Claude and Gemini?
It differs by model and plan and changes with every release. Check each tool's official models page.
Does memory use up the context window?
Yes, what is retrieved goes into the window. That is why retrieving only the right fact matters, not everything saved.
Is Memory Bridge a form of RAG?
It is a memory: notes that build up, get updated and are written by your tools, searched by meaning on each request.
Keep reading
Why ChatGPT forgets
ChatGPT forgets for three reasons: a new chat starts empty, a long chat outgrows the context window, and its memory is limited. Each cause and what to do about it.
AI memory glossary
Short, plain definitions of AI memory terms: context, context window, built-in and shared memory, semantic search, RAG and others.
One memory. Every tool you use.
The trial is invite-only for now. Request access and we will write to you when your seat opens.