Introducing the Conversation Compaction Option

Long conversations now have a cleaner way to stay useful without carrying full history on every turn. The new Compaction option in model settings gives teams direct control over how conversation context is reduced when thresholds are reached.

With compaction enabled, the platform can preserve continuity by rolling earlier turns into checkpoint-style context instead of relying only on hard truncation. This helps assistants stay grounded in prior discussion while keeping token pressure under control as conversations grow.

The option is available through the Language Model selector under Threshold Strategy, where you can choose Compact for summarization-driven history management or keep Truncate for latest-turn prioritization. Combined with Max Tokens and Interaction Max Messages, teams can tune memory depth and response stability to match support, operations, and multi-step assistant workflows.

This release gives builders a more practical balance between context quality and efficiency, especially in production conversations that extend over many turns.