RAG & Agents Engineering

M14 lab, vector stores and index operations

The record manager decides what gets embedded again.

Six Larkspur Cloud handbook files, eleven chunks. Change the folder, pick a cleanup mode and run index(): a JavaScript port of LangChain’s indexing logic, checked against langchain_core. The IDs are the real ones LangChain derives from each chunk’s text and source.

Cleanup mode for the next run

Source folder: what the loader returns next run

No run yet.

Embedded this run
0
Skipped, hash unchanged
0
Deleted this run
0
Stale chunks in the store
0

Vector store

Empty

IDSourceChunkThis run

A stale chunk is still in the store but no longer in the folder, so retrieval can still return it. Struck-through rows were deleted by the run that just finished.

Runs so far

RuncleanupAddedSkippedDeletedIn store