How to Set Up AI Memories and Folders in MultiChats

If you keep retyping the same background into every new chat, you are doing manual work the app can do for you. This guide shows you how to set up AI memories and folders in MultiChats so the model already knows your context the moment you start typing. We will build a real, tidy setup: a few general memories, a Work folder that carries its own memories, and a clear rule for which facts live where. It is written for people who already chat with AI daily and want a clean workspace instead of a scrap pile of repeated prompts.

One note up front: memories are a paid feature, available on any subscription, and that includes the memories you attach to a folder. Folders themselves are available for organizing chats without a subscription, so you can build the folder side of this setup on the free plan; the memory steps need Pro or Pro+. Pricing is at the end.

The three layers of memory

Before you touch a button, it helps to know how memory is structured. Rather than one big bucket, MultiChats merges three layers into the model's context for each chat, and knowing the layers is what keeps your setup clean instead of cluttered.

General (default): memories you mark as default. These ride along in every new chat automatically. Use this layer for stable facts about you that almost always apply: your name, your writing style, your timezone, the tools you live in.
Per-folder: memories attached to a folder. Any chat you create inside that folder inherits them. This is the layer most people underuse. A Work folder can carry your company name, product details, and house style without polluting your personal chats.
Per-chat: memories tied to a single conversation, including the short summary you can save when a chat gets long. This is context that matters here and nowhere else.

Why split them at all? Because everything you add becomes input tokens the model has to read on every turn. That input shares the same budget as the answer, the so-called context window. Stuffing your entire life into general memory means the model re-reads irrelevant facts in every chat, which wastes room and can dilute the answer. Layering keeps each chat carrying only what it needs.

A simple rule sorts most facts: if a fact is true no matter what you are doing, it is general; if it is true only for one job or subject, it is per-folder; if it is true only for one conversation, it is per-chat. When you are unsure, default to the narrower layer. You can always promote a memory upward later, and a tight general layer ages much better than a bloated one.

Step-by-step: create your first memory

Start with one general memory so you can see the layer working. On the web, memories live in settings; on mobile they live in a sheet. The flow is the same.

Open Settings and go to the Memories tab. On mobile, open the memories sheet from your profile.
Fill the Title field with a short label, for example "Who I am", and put the actual fact in the Content field. Keep content tight: one or two crisp sentences beat a paragraph.
Decide on scope with the Enabled by default toggle, described in the app as "Include in all new chats automatically". Turn it on for a fact you always want present. Leave it off for something you will attach selectively.
Click Add memory. That is it. Open a fresh chat and the model now has that fact without you typing a word.

Good first general memories: your name and role, your preferred tone ("plain English, no filler"), the stack or tools you use most, and any hard rule you repeat ("always give code in TypeScript"). Three or four of these covers most of what you keep retyping.

Step-by-step: set up a folder with its own memories

Now the part that makes per-folder AI memory worth it. A folder files your chats, and it does one more thing: each folder can carry its own selected memories, and chats inside it inherit them. Here is the build.

Create a folder. On web, use the new-folder control in the sidebar; on mobile, open Create folder and give it a Folder name. Name it for the context, not the project, so it stays reusable: "Work", "Client Acme", "Studies".
Open that folder's memory selector (the folder memories control) and pick which memories belong to this folder. These are the facts that should apply inside the folder but stay out of your other chats.
Drag existing chats into the folder, or start a new chat from inside it. Either way, the chat now inherits the folder's memories on top of your general defaults.
Optional: nest a child folder for a sub-context. A child folder copies the parent's memories when you do not pick its own, so a "Client Acme" folder inside "Work" starts with your work context already attached.

There is a second, faster way to feed a folder. When a conversation gets long, MultiChats offers to summarize it into a memory. If that chat is inside a folder, the save dialog gives you the scope Save to this folder, so the summary lands exactly where you want it. More on that flow in the guide on handling long AI chats.

Which scope for which job

When you save a memory out of the summarize flow, you choose a scope. Match the scope to how widely the fact applies. The table is your quick reference.

Scope option	Where it lands	Use it when
`Use in all chats`	General (default) layer	The fact is true everywhere: your name, tone, hard rules.
`Save to this folder`	The current folder's memories	It only matters for this context, e.g. one client or project. Shows only when the chat is in a folder.
`Just save to my memories`	Your library, not auto-applied	You want to keep it but attach it by hand later.

A starter setup you can copy

Here is a concrete setup that works for most people who use AI for both work and life. Build this once and most of your retyping disappears.

Three general memories (all with Enabled by default on):

"Who I am": your name, role, and timezone.
"How I like answers": tone and format, for example "plain English, short paragraphs, no filler, code in TypeScript".
"My stack": the tools and frameworks you ask about most.

One Work folder with its own memories (default off, attached to the folder):

"Company": your employer, what the product does, who the audience is.
"House style": brand voice rules, banned words, the way you format docs.

The payoff is clean separation. Ask a work question inside the Work folder and the model already knows your company and house style. Start a personal chat outside it and that work context stays out of the way. If you juggle several models, this context follows you when you switch models mid-chat, since memories attach to the conversation, not the model.

Setup checklist

Run down this list. If every line is true, your memories and folders are set up well.

You have three or four general memories with Enabled by default on, and nothing project-specific in them.
At least one folder exists, named for a reusable context, with its own memories attached.
Each memory is one or two sentences, not a wall of text.
You opened a fresh chat in the folder and confirmed the model knows the folder context without being told.
When a long chat is worth saving, you summarize it into a memory with the right scope instead of starting cold.

Keep your memories tidy over time

A memory setup rots if you never weed it. Treat the list like a small drawer you tidy now and then, not a junk pile that grows forever. Three habits keep it healthy.

Prune stale facts. If you changed jobs or dropped a tool, edit or remove the memory rather than letting the model carry a half-true fact into every chat. Memories you delete go to trash, so you can restore one if you cut it by mistake.
Watch what you mark default. Every default memory rides along in all new chats. If the model starts referencing context that does not belong, the usual culprit is a default memory that should be per-folder instead.
Merge duplicates. If two memories say nearly the same thing, combine them into one clean line. Fewer, sharper memories read better than many overlapping ones.

If the AI ever ignores a memory you expected it to use, check two things: whether the chat is in the folder you think it is, and whether the memory's default toggle is actually on. Most "it forgot" moments come down to a scope mismatch. Folders can also be nested and reordered, so if a chat sits in a child folder it inherits that child's memories, which may differ from the parent you set up.

Frequently asked questions

How does memory work in an AI chat?

A memory is a short fact you save once and reuse. MultiChats merges three layers into each chat's context: your general (default) memories, the memories attached to the folder the chat lives in, and anything saved to that single conversation. The model reads all of it as input, so you do not have to repeat yourself. For the bigger picture on saved context, see the explainer on AI chat with memory and folders.

What is the difference between general and per-folder memory?

General memory applies to every new chat once you flip its Enabled by default toggle on. Per-folder memory applies only to chats inside that folder. Use general for facts that are always true about you, and per-folder for context that belongs to one job, client, or subject. The split keeps each chat from carrying clutter it does not need.

Are memories and folders free?

Memories are the paid part and need an active subscription, Pro or Pro+. Pro includes up to 200 saved memories and Pro+ has no cap. Folders themselves are available without a subscription, so you can organize your chats into folders on the free plan. Pro is $20.99 a month and Pro+ is $34.99 a month, with yearly options that lower the monthly cost.

How do I make the AI remember something in every chat?

Create the memory in the Memories tab and turn on Enabled by default before you press Add memory. If you are saving from a long chat instead, pick the Use in all chats scope. Either route puts the fact in the general layer so every new chat starts with it.

Wrap up

Set up once, save time on every chat after. Keep general memories small and stable, push context-specific facts into folders, and let the summarize flow feed the right layer when a conversation has earned a keepsake. A tidy memory setup means the model meets you halfway every time you open a new chat. If you are ready to build it, see the plans to turn on memories; folders are ready for you on any plan.