session-summarize

Session summarization harvests validated SQL queries, business rules, and metric definitions from completed chat transcripts to prevent context loss. It automatically scans the session, classifies each valuable takeaway, and routes content to the appropriate store for persistence.

Persisting business rules and user preferences is achieved by harvesting high-value content from the chat transcript at the end of a session. The session summarization process collects durable preferences, classifies them using a decision tree, and routes them to designated stores.

Yes, a confirmation-first persistence mechanism presents a deduplicated Summary Manifest of all planned storage actions before any heavy generation or file writes occur. This avoids unnecessary work and unintended changes by requiring your explicit approval.

Storage classification uses a decision tree to route each harvested candidate to the correct store, such as semantic models, metrics, reference SQL, knowledge base, memory, or AGENTS.md. It assigns the appropriate generation mechanism based on the content type.

You need session summarization at the end of an interactive data exploration workflow where context from SQL validation, metric refinement, and business rule definition must be captured. It prevents the loss of validated multi-step workflows and business facts for future interactions.

Yes, session summarization deduplicates harvested candidates before routing them. It scans the transcript, removes redundant takeaways, and generates a clean Summary Manifest containing only unique persistence actions for your final review.