remember

Behavioral friction and lessons are captured as structured observations using explicit, contextual, or mining modes. The system extracts these data points from user input, ongoing conversations, or transcripts, storing them locally with tags for later synthesis via the rethink loop.

Session mining mode enables bulk extraction of behavioral patterns from transcripts. It categorizes extracted observations with tags and applies deduplication before storing them locally, ensuring clean data for later synthesis via the rethink loop.

Contextual correction detection scans ongoing conversations to identify behavioral friction and corrections as they occur. It captures these contextual patterns as structured observations, categorizes them with tags, and stores them locally to accumulate evidence for system improvements.

Yes, a local observation store is required. The workflow depends on this local store to accumulate categorized observations, apply deduplication, and manage the evidence needed to escalate ready categories to the rethink phase for system improvements.

The workflow monitors the local observation store for categories that have accumulated sufficient evidence. Once a category is ready, the workflow triggers the escalation to /rethink for downstream synthesis and system improvements.