AI agents · eve · agent memory

Agent memory in eve 1.0: per user or per team?

eve 1.0 lets you scope an agent's memory to one person or a whole team. The framework makes the split easy. Deciding what goes where is still on you.

Robert MillNovember 17, 2026 · 3 min read

A small personal notebook beside a large shared ledger on a wooden desk, illustrating memory kept by one person versus memory shared by a team

eve 1.0 shipped this week, and the feature I care about most is a small one: an agent's memory now has a scope. You decide, in one line, whether a memory belongs to one person or to everyone in a shared space. That is the right question to put in front of a builder, and it is the one I wrote about earlier this month in

The framework now makes the split easy. It does not make the decision for you.

What did eve 1.0 add?

covers a lot: multiplayer sessions, multi-tenant agents, tracing, faster turns. Memory gets a short section. Agents can carry context from one session to the next, and you choose who shares it.

The show how. Each memory is a "slot", one file in your agent's folder. Every slot has a scope. Use the built-in byPrincipal and each signed-in person gets their own memory. Return the id of a shared space, like a Slack channel, and everyone in that channel shares one. You can have both at once, as two slots that stay separate.

That maps closely onto the two layers I argued for: personal context that belongs to one person, and shared context that belongs to the team.

Does that settle where context should live?

No. It settles where it can live. A scope is a container. What goes in it is still a judgment call.

My rule of thumb has not changed. How someone likes to work, their drafting style, the threads they are juggling: personal slot. Decisions, customer facts, the reason we did it this way: shared slot.

The mistake is forcing one layer to do the other's job, and a config option does not stop you from making it. An agent with only a per-person slot will still bury team decisions in one person's memory. An agent with only a shared slot will never learn how you think.

Which memory wins when they disagree?

This was my first rule: decide which one wins, and it should be the shared record.

From what I read, eve recalls every slot before the model runs and labels each piece with the slot it came from. I did not find anything that ranks one slot over another. So the tie-break is yours to write, in the agent's instructions: when personal memory and team memory disagree, trust the team's, and say so.

It is one sentence. It is also the sentence most people will skip. The result is what I described before: the most confident source takes over, and that is usually the wrong one.

Is the shared layer easy to write to?

My second rule was that adding a decision should take one sentence from the agent in the middle of a conversation. Here eve does well. With the built-in file memory, the model saves and removes entries itself through tools, so "remember that we dropped this feature" can go straight into the team's slot.

The catch is the other half of the shared layer's problem: it goes stale, and people trust what is written. A tool that makes writing easy makes stale entries easy too. A shared slot still needs an owner, someone who reads it and removes what is no longer true. No framework gives you that person.

What I would do with it

Two slots, not one. A personal slot scoped to the person, and a team slot scoped to the shared space. One line in the instructions saying the team slot wins. One named owner for the team slot.

Then, for each thing the agent wants to remember, ask the question I ended on last time: who needs to know this, and how long should they be able to trust it?

Agent memory in eve 1.0: per user or per team? · Robert Mill