The context window is the amount of text a language model can process at once, measured in tokens. It is the AI's working memory: big, but fleeting, expensive and easily distracted. An AI memory adds long-term memory on top, bringing only what matters into the context.
What is a context window?
The context window is the amount of text a language model has "in view" at once. That includes your messages, the AI's replies, attached files and instructions. It is measured in tokens, the small chunks of text the model computes with.
The context windows of ChatGPT, Claude and Gemini have grown enormously in recent years, from a few thousand to around one million tokens today, and up to ten million in some models. That sounds like the memory problem is solved. It isn't.
Working memory, not long-term memory
Psychology distinguishes between working memory and long-term memory. Working memory keeps present what we are working on right now. Long-term memory keeps what should last and retrieves only what fits. An AI's context window is working memory. No working memory, however large, replaces long-term memory.
Three reasons AI forgets despite a large context window
1. Fleeting
When the session ends, the contents of the context window are gone. The next session starts from zero. Even within long chats, the oldest material eventually drops out or gets summarized, and details are lost. Hence the familiar feeling: "ChatGPT forgets things."
2. Expensive
Every token in the context is processed again with every request. If you send long notes or memory files with each message, you pay for them every time, in money, waiting time or subscription quota.
3. Distracting
More context doesn't make models better, it makes them worse. Language models are clearly distracted by irrelevant information (Shi et al., ICML 2023), and their performance drops with every additional token. What the research shows is covered in the guide Less context, smarter agent.
It's not how much an AI can see that counts, but what it sees.
How an AI memory relieves the context window
Giving an agent exactly the right information, not too little and not too much, is called context engineering. An AI memory like Hebby is context engineering on autopilot. Instead of sending everything along, Hebby brings only the memories that matter now into the context, plus the ones you're about to need. A purpose-built graph memory makes sure the connections come along too. That keeps the context window free for the actual work, and your agent still forgets nothing important.
Hebby Memories: the second brain for Claude Code, Codex & co. Learns as you work, thinks ahead, never forgets what matters. Get Early Access →
Frequently asked questions
What is a context window?
The context window is the maximum amount of text a language model can consider in a single request, including your messages, its replies and any attached documents. It is measured in tokens.
What is a token?
A token is a chunk of text that language models compute with. Roughly, one token is a word fragment, in English usually a bit less than one word.
How big is the context window of ChatGPT, Claude and Gemini?
Today's leading models process around one million tokens at once, some up to ten million. How much of that you can use depends on the model and the plan. Only a fraction is used well anyway.