Concept·AI Models & Capabilities·Added 1 day ago

Context

Also known as: model context, prompt context, conversation context, context length, the context

Everything the model can see when it generates a response: your prompt, the conversation history, any documents or data you've included, tool results, and the system prompt. The model has no memory outside this window. 'Context' is the whole input universe the model reasons over.

Context is the most overloaded term in AI builder discourse. At its broadest, it means everything passed to the model in a single call: your message, previous messages in the thread, any files or data you attached, tool call results, and the system prompt (the background instructions set by the developer). The model generates its response based only on what's in this context. There is nothing else.

This has major practical implications. If important information isn't in the context, the model doesn't know it. If the context gets very long, the model may lose track of earlier parts (the 'lost in the middle' problem). If the same context is reused many times, it can be cached to save costs. If the context fills up before a task is done (context rot or context compaction), the agent may lose crucial information.

Context engineering is the emerging practice of deliberately managing what goes into this window: what you include, what you summarize, what order you put things in, and what you leave out. It's one of the highest-leverage skills in building with AI right now, because the model's entire world is just what you put in front of it.

This definition is AI-generated and refreshed weekly. It may contain inaccuracies. Use your own judgment, especially for production decisions.

Related terms