Skip to content

Give your agent a memory that survives tomorrow

Why does an agent forget everything? Memory is a file of short notes it reads before every job and writes after. The smallest version that works.

By Nick

2 min read

AI AgentsPaste This Intermediate
On this page
Time
Twenty minutes
Cost
Nothing extra
Level
Intermediate

Before you start

1. What memory actually is

Every run starts with a clear desk. Memory is the notebook you leave on it: a plain file the agent reads first and writes last. “Nick prefers short answers.” “The invoice job needs the client folder, not the whole drive.” “The calendar sync moved on 3 October; check before trusting it.”

Not transcripts. Not everything. Facts that would change how the next job goes.

2. Set it up with one instruction

Paste this into the agent, and keep the file it creates somewhere you can open.

Set up a long-term memory for yourself: one plain text file. Before every job, read it.
After every job, add any new fact that would change how a future job should go: a
preference I stated, a correction I made, something about the environment that changed.
One fact per line, dated, under 20 words, no transcripts, no summaries of what you did.
Never delete a line without asking me. Tell me where the file lives, and show me its
contents whenever I ask.

3. Back it up before an agent writes to it

Before you let anything write to this file automatically, keep a copy you can roll back to: a version-controlled folder, a synced backup, or just a dated copy saved before the first automatic write. A bad edit to a file an agent reads before every job is a mistake that repeats itself until you notice it.

4. The rule that stops it rotting

Once a month, read it. Delete the stale, merge the duplicates, fix the wrong. Ten minutes. The weekly check includes “what did you add to memory”, so you see it drift before it matters.

And keep one memory, not one per bot. Four bots with four notebooks is four bots that each know a quarter of what you taught them; I did that, and collapsed them back into one.

How you know it worked. You can open the file and read it in two minutes. Honestly, this is still the weak part of my own setup: it works, but it is more manual than it should be.

Common questions

Is this the same as the memory in ChatGPT?

Same idea, your copy. The apps keep a small notebook about you behind the scenes. An agent you run keeps one you can open, edit and delete, which is better and also your job to look after.

Why not just keep every conversation?

Because it all gets sent to the model each time, which costs tokens and buries the facts that matter. A memory full of everything is as useless as no memory. Short, specific notes beat transcripts.

One memory or one per bot?

One, unless isolation is the point. I ran four bots that each knew a quarter of what I had taught them and collapsed them into one. The Grok Bot comparison has the long version.

What to do next

Related reading

Last reviewed 9 October 2026 by Nick.

Software changes fast. If something here is out of date, tell me and I'll fix it.

Get new guides by email

Occasional and practical. No spam, no hype, unsubscribe whenever.

Occasional practical guides from UncomplicateAI. You will get a confirmation email before anything else is sent. Unsubscribe any time. Privacy

You are not subscribed until you confirm the email. Unsubscribe any time.