# Strands Context Management Context management is how [[Strands Agents]] keeps long conversations within the token budget and avoids the performance degradation that comes from overstuffed context. It offers two modes: - **Automatic** (`"auto"`, default), a `SummarizingConversationManager` plus a `ContextOffloader` proactively summarize history and offload large tool outputs; a `SlidingWindowConversationManager` can drop old messages to prevent overflow - **Agentic** (`"agentic"`), hands control to the model via tools like `summarize_context`, `truncate_context`, and `pin_context` This is a production concern that distinguishes mature agent frameworks from naive loops, and it works hand-in-hand with [[Strands State|state]] and the [[Context Window]]. ## Related - [[Strands Agents]] - [[Strands State]] - [[Context Window]] - [[Agentic Engineering]]