recall

To find past context from previous AI sessions, use a breadth-first search to query multiple persistent artifacts like beads, memory keys, and topic files. This cross-corpus discovery returns lightweight, ranked previews of past discussions matching your vague topic reference.

Retrieving knowledge from memory artifacts in a cold-start conversation requires parallel multi-source querying across persistent storage. The system applies recency-aware ranking and deduplication to provide preview-only outputs, ensuring efficient context discovery without automatically pulling full content.

Recency-aware ranking prioritizes recently touched items and cross-source matches during context discovery. This ranking mechanism improves retrieval speed for past discussions by surfacing the most relevant persistent memory artifacts first, avoiding unnecessary deep reads of outdated files.

Yes, you can search for named entities and project references without loaded context by querying persistent artifacts like beads and topic files. The retrieval process handles vague references by returning ranked one-line previews instead of pulling full content too early.

A limitation of using breadth-first search for memory retrieval is that it returns preview-only output without automatic deep reads. This keeps context usage efficient but means you must manually identify and select the correct past discussion from the ranked one-line previews before accessing full content.