A context window is the maximum number of tokens — roughly, chunks of text — a model can take into account in a single request. Everything the model 'sees' (your instructions, the code, prior conversation, tool output) must fit within it.
Context windows shape agent workflows: larger windows let an agent reason over more of a codebase at once, while smaller ones require careful selection and handoffs to preserve important context between sessions.
Learn more
Related guides on the Agents UI blog and related terms in this glossary.