The token capacity available for the information a model processes in an interaction, including relevant input context and applicable output limits; not permanent memory.
A context window is the capacity available for the information a model processes during an interaction. It is usually expressed in tokens. Depending on the system, that information can include instructions, conversation history, retrieved material and room for generated output. It is not the same as permanent memory of everything a user has ever said.
A larger capacity allows more material to be supplied, but it does not guarantee that every detail will be used correctly. Relevance, organization and the clarity of the task still matter. An application may also summarize, select or omit older material, so the full visible conversation is not always identical to the context used for a particular response.
For a business workflow, important requirements should be preserved deliberately rather than assumed to remain available forever. Keep authoritative source documents and decisions outside a chat transcript, and provide the relevant pieces when needed. A model’s confident answer is not proof that it retained a restriction mentioned much earlier.
Takeaway: Manage context as a limited working set. Supply the evidence and constraints needed for the task, and verify that important requirements survive long conversations or document-processing steps.
Example
A team reviews a long product catalogue with an AI assistant. It keeps a separate approved specification file and passes the relevant section for each comparison. That is more reliable than assuming an early correction remains fully available after many unrelated exchanges.
What to check
Check the current model and application limits before a large task. Ask how excess input, older messages and retrieved documents are handled. Reserve adequate output capacity and validate the final result against the authoritative source rather than relying on apparent conversational memory.
For more detail, see OpenAI token and limit documentation.