Automatic Context Compression

A key mechanism for controlling token costs in long-conversation agents. It employs summarization, truncation, or selective retention strategies to condense earlier dialogue into brief memories, retaining only information strongly relevant to the current task.