Atlas · models
Context window
Everything the model can see on one turn, with a hard size limit measured in tokens.
See it working: Plate I
In plain words
The model has no memory between calls. On each turn it sees only what is in its context window: instructions, the conversation so far, and tool results.
How it works
The window has a fixed budget of tokens, and the reply must fit in it too. Context engineering is the craft of choosing what goes in: enough to do the task, little enough to leave room.
Going deeper
Plate I uses a toy window of 2,000 tokens so you can watch it fill. Real windows are far larger, but long agent sessions still reach them.
Context is only text. Simon Willison’s files-to-prompt turns a folder into one prompt you can read before any model sees it. Codex ran it on a two-file folder on 10 November 2026; with -e md only the Markdown file was printed:
fixture/readme.md
---
# R-001
Markdown context.
---
Sources
- Conversation state OpenAI
maximum number of tokens that can be used in a single request
Model inputs and outputs are metered in tokens
- Effective context engineering for AI agents Anthropic
Context refers to the set of tokens included when sampling from a large-language model (LLM).
the model’s ability to accurately recall information from that context decreases.
Context, therefore, must be treated as a finite resource with diminishing marginal returns.
This results in n² pairwise relationships for n tokens.
Compaction typically serves as the first lever in context engineering to drive better long-term coherence.
These notes get pulled back into the context window at later times.
but returns only a condensed, distilled summary of its work
- files-to-prompt Simon Willison
Concatenate a directory full of files into a single prompt for use with LLMs
Quotes were checked against the live pages on 2026-11-08.