A memory strategy that condenses older messages into an AI-generated summary.
When the message count exceeds the threshold, the oldest messages are summarized into a single pinned system message, while the most recent messages are kept intact.
Options
:threshold- minimum message count before summarization kicks in (default: 20):summarize_fn- a 3-arity function for testing (avoids real AI calls):provider- AI provider override (falls back to context):model- AI model override (falls back to context)
Priority
Returns 300.