Join the waitlist to take part in the alpha.

Your agents' memory

What your agents remember from one conversation to the next, how that memory is built, and how you stay in control of it.

An agent without memory starts from scratch at every question. You explain again who the client is, what the procedure is, what you decided last week.

Standards agents remember. Not the conversation word for word, but what counts: the context, the decisions, the way you work.

Two memories, two jobs

Conversation memory. What was said in the current session. When the exchange gets long, it is compacted: older turns are summarized to free up room, while the decisions taken are preserved. You do not lose the thread, and neither does the agent.

Long-term memory. What the agent has learned about your company over many sessions, and finds again in a completely unrelated conversation. This is the memory that makes an agent useful after a few weeks.

How a memory is found again

Memory is not a list that gets read from end to end. Each memorized item is indexed by meaning, not by exact words. When you ask a question, the agent goes looking for what comes close and pulls it into its working context.

In practice: you do not need to phrase it the way you did the first time. A question asked differently finds the same memory.

Staying in control

Memory is a tool, not a black box. Three principles hold:

  • An agent only memorizes what it was allowed to read.
  • What is memorized can be inspected, not guessed at.
  • Out-of-date information can be corrected: the new version is what counts from then on.

Frequently asked questions

Can an agent remember something I do not have access to?
What an agent memorizes comes from what it was able to read, under its own rights. Those do not necessarily overlap with yours, either way.
What happens when a conversation gets very long?
It gets compacted. Older exchanges are summarized to make room for new ones, keeping the decisions and the useful context.
Was this article helpful?
Go further
Choosing the model your agents use

Your workspace is tied to no AI provider. How the model is chosen, what happens if it goes down, and what it costs.

Can't find it?

Write to us and someone who knows the product will answer.

Contact us