context-optimization

Context optimization extends effective context capacity through strategic compression, masking, caching, and partitioning. It doubles or triples available capacity without requiring larger models or longer context windows.

Context partitioning divides context across multi-agent workflows to improve efficiency. You need it when processing large inputs in hierarchical planning or running long-running agents that exceed standard context limits.

Summarize the current context and replace verbose outputs with compact references. Context compaction reduces token usage while preserving essential information for continued processing.

Observation masking works with large document processing by selectively hiding irrelevant observations. It ensures the framework handles large inputs efficiently without expanding the model context.

Context compression does not increase actual context window size. It optimizes available capacity through compaction and partitioning, meaning extremely large inputs may still require architectural changes.

Context budgeting and monitoring integrate through a memory system interface. They track and allocate token usage across long-running agents, ensuring efficient context utilization.