Guides / How Esment works.

1329 words | Esment

How Esment works.

Esment gives Claude, ChatGPT, Cursor and other assistants one shared memory. What gets saved, what they read first, and where it lives.

╌╌╌╌

Every assistant you use keeps its own memory, if it keeps one at all. ChatGPT does not know what you told Claude, Cursor does not know what you decided in Claude Code, and a new chat in any of them usually starts from nothing. Esment is a memory layer that sits underneath all of them: one place where what matters about you and your work is kept, which every connected assistant can read and add to.

Esment is our product, so read this as a description from the people who build it. It is not an assistant and not a model. It does not answer questions. It remembers, and hands the right things to whichever assistant you are talking to.

One Esment memory in the middle; desktop assistants connect to it directly, web assistants through an addressON YOUR MACCONNECTED DIRECTLYON THE WEBTHROUGH AN ADDRESSCLAUDE DESKTOPCLAUDE CODECURSORWINDSURFCODEXRAYCASTADDRESSCHATGPTCLAUDE.AIPERPLEXITYGROKTHE ADDRESS IS ESMENT CLOUD, OR A TUNNEL TO YOUR MACONE MEMORY

One memory, many assistants

Assistants connect to Esment over MCP, the open protocol they already use to reach outside tools. Through it, an assistant gets a small set of tools: save a memory, search memory, read the brief. Because every assistant talks to the same store, a fact saved from one is there for all the others.

How you connect depends on where the assistant runs. Desktop apps on your Mac, such as Claude Desktop, Claude Code, Cursor, Windsurf, Codex and Raycast, are set up from the Esment app with one click on Configure. Quit and reopen the assistant afterwards; closing its window is not enough. See connecting your assistants.

Web assistants, such as ChatGPT, claude.ai, Perplexity and Grok, run on their makers’ servers and cannot see your Mac. They need an address on the internet. That address is either Esment Cloud, or a temporary tunnel the Mac app opens to your computer through Cloudflare. The web assistants page explains the difference.

How memories get saved

A memory is one small thing worth keeping: a fact, a preference, a decision, a piece of context. There are three ways one gets saved, and you can mix them freely.

  • Automatically. While you chat with a connected assistant, it saves durable details it notices, such as your role, a preference or a decision.
  • Because you asked. Say “remember that I prefer short replies” and the assistant saves it straight away. “Forget that” removes it.
  • By you. Add or edit memories yourself in the app or the web dashboard.

Each memory records when it was saved, what kind it is (a Fact, Knowledge, a Moment or a How-to), its tags, and which assistant saved it, shown as “via ChatGPT” or “via Claude”. If something looks wrong, that tells you where it came from. Anything you see can be edited or forgotten.

What an assistant reads first

Connecting a memory is not the same as using it. An assistant that has to decide for itself when to look something up often does not think to. So Esment does not wait to be asked. It prepares a short session brief in advance: the people, projects and preferences that matter to you, your pinned memories, your Memory Blocks and what you worked on recently.

Before a chat starts, a short brief is prepared from your memory; the assistant reads it first and can search for more laterYOUR MEMORYSESSION BRIEFPEOPLEPROJECTSPREFERENCESPINNEDMEMORY BLOCKSRECENT WORKFIRSTNEW CHATBRIEF, ALREADY READLATER IN THE CHAT, IT CAN SEARCH YOUR MEMORY FOR A SPECIFIC FACT

A connected assistant reads the brief before it answers, so a new chat starts from what you already know instead of from nothing. Later in the conversation it can also search your memory for something specific. Some assistants, ChatGPT among them, need to be told once to work this way.

The brief exists because of a limit every model has: it can only read so much at once. Pasting all of your memory into every chat would crowd out the conversation, cost more, and make the model worse at finding what matters. The context window chapter explains why.

How search finds the right memory

Search works by meaning, not by matching words. Every memory is turned into an embedding, a list of numbers that places it in a space where similar meanings sit close together. A question is turned into one too, and the memories nearest to it come back. That is how “what do I drink in the morning?” finds “oat milk flat white, no sugar”.

On the Mac, this runs on your computer: nothing you write is sent anywhere to find the right memory, and it works across English, Spanish, German, French and Italian, even when you saved something in one language and ask in another.

Summaries that stay current

Besides the memories themselves, Esment writes a short summary for the people, projects and topics that come up most, labelled Auto summary so you can tell it apart from what you or your assistants saved. Each one keeps a single current version, rewritten only when there is real news to add.

The life of a memory

A memory that never changes state eventually becomes clutter. So each one has a state, shown as a label, and most changes happen on their own.

A memory is saved as new, becomes active, and may later be merged, archived and restored, or forgotten+NEWJUST SAVEDACTIVEIN REGULAR USEMERGEDJOINED WITH A TWINSET ASIDERESTOREARCHIVEDSTILL KEPTFORGOTTENNO LONGER OFFEREDPINNED MEMORIES ARE NEVER ARCHIVED AUTOMATICALLY
  • New: just saved.
  • Active: in regular use and available to your assistants.
  • Merged: combined with a related memory, so the same thing is not stored twice.
  • Archived: set aside as it becomes less relevant, but still kept. Restore brings it back.
  • Forgotten: no longer offered to assistants, because you forgot it or because it aged out.

Pin a memory when it must never fade, such as an allergy or the name of your company. A pinned memory is never archived automatically and is always part of the brief. The Mac app calls this Evergreen. The brief has limited room, so pin sparingly. Find, edit and forget covers the controls.

Where your memory lives

Mac app, sync offWeb and Esment Cloud
Where memories are storedA hidden .esment folder in your home folderEsment’s servers in the United States, in your workspace
Where search runsOn your Mac, offlineHosted: OpenAI computes embeddings and Voyage AI ranks results
Who reads what is foundThe assistant you connected, under its own policyThe assistant you connected, under its own policy

The last row is the same in both columns, and it is worth being clear about. Whatever Esment hands to an assistant is read by that assistant’s model, which usually runs on its maker’s servers. Esment controls where your memory is stored and searched; it does not control what ChatGPT or Claude do with what they read. Our privacy page lists everything that can leave the Mac.

What it does not do

  • It cannot force an assistant to use it. The brief helps a great deal, but each assistant still decides when to save and when to search.
  • It does not make a model smarter about things it was never told. It remembers what you or your assistants saved, nothing more.
  • The Mac app needs a paid plan and a network connection to check your licence. The Free plan is web only.
  • It is not open source. Workspaces shared by several people come with the Team and Enterprise plans; see Teams.

Save your first memory →

Common questions

In:

Does Esment replace the memory built into ChatGPT or Claude?

Out:

No. You can keep both. The built-in memory stays inside one assistant; Esment is shared across all the ones you connect, and you can see, edit and forget everything in it.

In:

Do I have to tell the assistant to remember things?

Out:

No. Connected assistants save durable details on their own while you chat, and read the session brief at the start of each conversation. Saying “remember that…” is there for when you want to be sure.

In:

Can I see which assistant saved a memory?

Out:

Yes. Memories saved by an assistant show its name, for example “via ChatGPT” or “via Claude”. Memories you added yourself show no name.

In:

Does my memory leave my Mac?

Out:

With the Mac app and sync off, memories are stored and searched on your Mac. What Esment hands to an assistant is read by that assistant’s model, wherever it runs. Esment Cloud and the web dashboard store memory on Esment’s servers in the United States.

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.

0.00