How Compaction Works in Pi

When coding sessions with AI agents like Pi exceed the language model's context window limit, the system triggers "compaction" to compress the conversation history. Rather than discarding all prior context or starting fresh, compaction uses an LLM request to summarize the conversation, replacing parts of the history with a compressed representation to free up space for continued interaction. This approach preserves important prior decisions and work while maintaining the ability to continue the coding session.