Retrieval Backed Memory: Files vs Embedding Stores

Forge keeps its notes about one repository in a plain file of 300 lines. Should you build an embedding index over that file, so that Forge can find a note by its meaning instead of by its exact words?

Many published architecture diagrams for agent memory include a vector database, so the question often goes unasked. The answer depends on three measurable things: how large the memory is, how quickly it changes, and whether the words of a later question match the words that were stored.