Context Window
The maximum amount of text (measured in tokens) a language model can consider at once — its prompt plus everything it has generated so far.
The context window sets the Cost of Retrieval an agent pays: the cheaper it is to fit your content in-window (markdown twins, concise structured data), the more of it an agent can afford to read.
- term
- Context Window
- category
- knowledge-memory
- short_def
- The maximum amount of text (measured in tokens) a language model can consider at once — its prompt plus everything it has generated so far.
- long_def
- The context window is the model's working memory: everything the model can 'see' for a single response — system prompt, retrieved documents, conversation history and tools — must fit inside it. Larger windows let an agent hold more context, but cost and latency rise with how much of the window is used.
- see_also
token-economicstokenizationcontext-engineering- etymology_origin
- — verify-against-primary-at-build ↗ https://en.wikipedia.org/wiki/Large_language_model — 'context window' is standard LLM terminology; no single coiner
- related_to
token-economicstokenizationcontext-engineeringprompt-caching- contrast_with
- Unlike a database that stores unlimited data, the context window is a fixed per-request budget — anything beyond it must be retrieved, summarized or dropped.
- example
- A model with a 1M-token context window can read a large codebase in one request; a 200K window forces an agent to retrieve only the relevant files.
- source
- https://en.wikipedia.org/wiki/Large_language_model
- status
- active
- why_it_matters
- The context window sets the Cost of Retrieval an agent pays: the cheaper it is to fit your content in-window (markdown twins, concise structured data), the more of it an agent can afford to read.
- sameAs
https://en.wikipedia.org/wiki/Large_language_model- bridge_entity
- context-engineering
- last_verified
- 2026-07-06
- md_twin
- /glossary/context-window.md
last verified · by Özden Erdinc