1238 words | Esment
AI with memory.
Which AI assistants remember past conversations, how each one does it, why none remembers everything, and how to get one that remembers across tools.
╌╌╌╌
Most people looking for an AI that remembers want something simple: stop repeating themselves. Say once that you work in Lisbon, that the client is called Northwind, that you prefer short answers, and have the assistant know it next week. That is possible today, but “remembers” means very different things depending on the product, and the difference decides whether it will work for you.
The short version: a language model on its own remembers nothing between conversations. Every assistant that seems to remember has a separate memory built around the model, which decides what to keep and puts some of it back in front of the model later. The questions worth asking are what that memory keeps, who chooses what comes back, and where it lives.
Why an AI forgets by default
A model reads text and writes text. What it seems to know in a conversation is whatever is in its context window at that moment: your messages, its replies, any files you added. When the conversation ends, none of it is written back into the model. The next conversation starts empty unless something outside the model fills it in.
So “AI with memory” really means an AI connected to a memory. That memory can be a short list of facts, a summary, a search over old chats, or a full store of its own. Each one remembers different things.
The four kinds of AI memory
| How it works | Remembers well | Forgets | |
|---|---|---|---|
| Saved facts | A short list of things you asked it to keep, or it chose to keep | Name, role, preferences, a few constants | Anything that did not make the list; the list fills up |
| Running summary | The assistant rewrites a summary of you as you talk | The general picture of who you are | Exact details, dates and which version of a fact is current |
| Chat history search | It searches old conversations when it thinks they matter | Things that were said, word for word | Anything it does not think to look for |
| A separate memory store | An external memory several assistants read and write | Individual facts, decisions and their history, across tools | Only what is never saved to it |
ChatGPT and Claude both combine the first three. Our guides on making ChatGPT remember and on whether Claude remembers across chats explain each one, with the settings to check. The fourth kind is what memory engines do, and the rest of this guide explains when you need one.
Is there an AI that remembers everything?
No, and you would not want one. Keeping every word of every conversation is easy; using it is the hard part. A memory that brings back everything brings back noise: the plan you dropped, the price that changed, the joke you made at midnight. What feels like “remembering everything” is a memory that keeps the facts that matter, finds them when they are needed, and lets the stale ones fade.
That is also why assistants with built-in memory sometimes seem to forget. They keep only what they judge important, they bring it back only when they judge it relevant, and they do not show you when a fact was replaced by a newer one. For personal preferences that is a good trade. For a project, a client or a body of research it is often not enough.
What to look for in an AI that remembers
- You can read it. A list or a store you can open, not a hidden profile. If you cannot see what it remembers, you cannot correct it.
- You can fix it. Edit a wrong fact, delete a private one, and have the change stick.
- It knows what changed. When you move cities or a decision is reversed, the old fact should become history, not a second answer.
- It does not depend on luck. Something should put the important facts in front of the model before it answers, instead of waiting for the model to decide to search.
- It works where you work. If you use ChatGPT for writing and Claude or Cursor for code, a memory locked inside one of them covers half your day.
- You know where it is stored. On your device, in the vendor’s cloud, or both, and what happens when you delete it.
One memory for every assistant
Built-in memory belongs to one app. ChatGPT’s memory is in your OpenAI account; Claude’s is in your Anthropic account; neither can read the other. The way around that is a memory outside both, which every assistant connects to over MCP, the open protocol most AI apps now support for tools. A fact saved while talking to ChatGPT is then there when you open Claude, and a decision made in Cursor is there in ChatGPT.
Our guide to sharing memory between ChatGPT and Claude walks through the setup step by step.
Where Esment fits
Esment is the memory we build, so read this part with that in mind. It is a separate memory that ChatGPT, Claude, Cursor, Claude Code and other assistants connect to. Each memory is stored as its own entry, with its type, its tags, when it was saved and which assistant saved it, and you can read, edit or delete every one of them in the app. Search works by meaning, so “where did I say the launch would be?” finds “we moved the launch to March”. When a fact changes, the earlier version is kept as history instead of overwritten. Before an assistant answers, it reads a short brief of what matters right now, so recall does not depend on the model choosing to search.
The Mac app keeps memory on your machine; the web app works from any browser. The Free plan holds up to 500 memories, and pricing lists the rest. If you have years of ChatGPT conversations, the Mac app can import them and pull out the facts, so you start with an AI that already knows you. How Esment works explains the design in more detail.
Connect your assistants to one memory →
Common questions
Which AI remembers previous conversations?
Out:ChatGPT and Claude both can, when memory is on: they keep saved facts or a summary about you and can look back at past chats. Each one remembers only inside its own app. To have several assistants remember the same things, connect them to a shared memory such as Esment.
Is there an AI chatbot with long-term memory?
Out:Yes. ChatGPT and Claude have built-in long-term memory for personal facts and preferences. For larger or changing knowledge, such as projects, clients or research, a separate memory store connected over MCP keeps individual facts with their dates and history.
Why does my AI forget what I told it?
Out:Usually because the fact was never saved, memory was off or the chat was temporary, the assistant did not judge it relevant to the new question, or its saved-memory list is full. Saving the fact explicitly and checking the memory settings fixes most cases.
Is it safe to give an AI memory?
Out:It depends on where the memory is stored and whether you control it. Prefer a memory you can read, edit and delete, check whether it stays on your device or goes to the vendor’s cloud, and keep passwords, ID numbers and financial details out of it.
Written by the team that builds Esment, so we are not a neutral party. Tools change quickly: check each one before you rely on it.
Get the next chapter by email.
One email per new chapter. Nothing else.