Which context management strategy is most appropriate for a long-runni — Agentic Systems Engineer Practice — Claude Cert Academy

Which context management strategy is most appropriate for a long-running research agent that must preserve specific facts discovered early in a multi-hour session?

• Rely on the model's full context window — modern models can hold multi-hour sessions
• Extract and store key facts in an external structured store; retrieve them on demand into the active context
• Use aggressive truncation to keep the context window small and latency low
• Summarize all earlier context into a single paragraph at each step

Answer: Extract and store key facts in an external structured store; retrieve them on demand into the active context

An external structured store (key-value store, database, or semantic search index) persists facts reliably regardless of context window size and allows precise retrieval of specific facts when needed. The full-context approach fails when the session exceeds the window limit. Aggressive truncation may discard the critical early facts. Single-paragraph summarization loses specificity — the exact wording, values, or citations of early facts may be altered in compression.

Continue to Claude Cert Academy