Manage the context window as working memory — what to include, summarise, or drop to keep agents coherent across long conversations.
Five passes over the same idea, each from a different angle. Do them in order, or jump to whichever you need.
The context window is an agent's short-term memory. It fills up fast. Strategies like sliding-window truncation, conversation summarisation, and message compression let agents maintain coherence without losing critical information. Understanding token budgets, what the model attends to, and how to structure context for retrieval is fundamental to long-running agents.
