Context window

Garden · not yet reviewed Established: widely used Widely discussed since 2023

The amount of material an AI can hold in view during one conversation. Anything beyond it gets dropped or summarised, and very long chats can make earlier details easier to lose.

Also called: context length · token limit · how much it can read at once · long context

Connections

Understand first

Ideas this one builds on.

• Large language model (garden)

Leads on to

Ideas that build on this one.

• Context engineering (garden)
• Memory (garden)

Resources that teach it

• AI Capabilities and Limitations (garden)
• The AI Agent Glossary (garden)
• n8n Tutorial: Build an AI Workflow to Extract Top Insights From All Your Newsletter Emails (garden)

Competency

• Describe the task

Find context window on the commons map ↗

You may know it from

• psychology working memory People can juggle only so much at once; past that, things slip.
• meetings the whiteboard in a meeting room Whatever gets erased to make space is gone unless someone wrote it down elsewhere.

In products

The same idea, under different names in different tools.

Provider	What they call it
Google	Gemini long-context models
Anthropic	Claude long-context models

In plain words

Each AI conversation has a limit on how much text it can consider at once, measured in tokens (roughly, pieces of words). Your messages, the AI’s replies, uploaded files and instructions all count toward it. Modern windows are large but finite.

Even well inside the limit, size has costs. In a very long conversation, details from the start can carry less weight, and a mistake made early can keep echoing. When a big file is uploaded, the tool may only pull in the parts it judges relevant.

Start a fresh chat when you switch topics. When a chat gets long and muddled, ask for a summary of decisions so far, then paste that summary into a new chat. Point to the exact section of a long document that matters.

Example

A program manager has been refining a project plan in one chat for two hours. The AI starts reintroducing a task she cut an hour ago. She asks for a clean summary of the current plan, opens a new chat with it, and the drift stops.

How it connects

• It sets the size of the box that context engineering fills.
• Memory and projects are ways to carry information across chats without stuffing one window.
• Lost details in long chats can look like hallucination.

Try it

Open a long chat you already have and jot down the decisions you think you’ve made in it. Then ask: “List the decisions we’ve made so far.” Note anything it forgot or invented.

---

Read this card as plain Markdown: /concepts/con-context-window.md. Know a parallel from your own field, or a product that implements this? Add it.