Definition

As conversations become longer, sending the entire conversation history to a model can become inefficient.

Context summarization reduces this problem by compressing older information into a shorter representation. For example, an application could:

  • Keep the most recent messages directly available.
  • Summarize older conversations.
  • Preserve important preferences or facts.
  • Send the resulting context to the larger model.

This can help control context size and inference costs while retaining useful information from previous interactions.