What is a context window?

Definition: A context window is the maximum amount of input and output, measured in tokens, that a language model can consider in a single request, including instructions, conversation history and any documents supplied.

How a context window works

Text is split into tokens, which are word pieces. Everything the model sees in one call, the system prompt, the user's message, earlier turns, retrieved documents and the model's own reply, must fit within the window. Anything beyond it is cut off or has to be summarized.

Why it matters

Larger windows let a model read long contracts, codebases or chat histories in one pass. But more context costs more, adds latency, and does not guarantee the model uses every part of it well; details buried in the middle of a long prompt can be missed.

How teams use it

Engineers manage context deliberately: they retrieve only the most relevant passages with RAG, summarize older conversation turns, keep instructions short and test answer quality as inputs grow. Prompt engineering and context design go hand in hand.

How Ryz helps

Ryz LLM engineers and RAG engineers design retrieval and context strategies that balance cost and accuracy, working on your team.

Related

• LLM development services
• RAG vs fine-tuning
• Large language model (LLM)
• Embeddings