AI companion with memory: which apps actually remember you (2026)

You tell an AI companion about your week, your sister's operation, the name of your dog. It listens beautifully. Then you open the app three days later and it greets you like a polite stranger. If that has happened to you, nothing was wrong with you or with how you used the app. Many AI companions simply do not keep much of you between conversations, and it helps to understand why before you invest months of talking in one.

We build a companion app, My Softly, and memory is the part we spent the most time on. So this guide does three things. It explains why the forgetting happens, it walks through how a memory system is actually put together (using ours, because it is the one we can describe from the inside), and it sets out how Replika, Nomi, Kindroid, Character.AI and Pi each describe their own memory, as of September 2026. There is also a simple test you can run yourself on any of them.

The short answer: memory design, app by app

Every app in this table says it remembers you. What differs is how. This is a comparison of design, taken from each company's own help pages and announcements as of September 2026. It is not a score, and we have not run a test on these apps. To see how any of them does for you, use our DIY memory test.

If you are choosing between these apps on more than memory (price, content policy, who texts first), our comparison of the best AI companion apps puts them side by side.

Why AI companions forget you

Every AI companion runs on a language model, and every language model has something called a context window. Think of it as the model's desk. Whatever is on the desk right now, it can see and respond to. Whatever has been cleared off the desk is gone, as if it never existed.

When you send a message, the app puts your recent conversation on that desk, along with instructions about the companion's personality. The model reads it all and writes a reply. But the desk has a fixed size. Once a conversation gets long, or once you start a new session, older messages get cleared away to make room. The model is not choosing to forget you. It never sees those messages again.

People call this the goldfish problem. Within one conversation the companion can seem wonderfully attentive, because everything you have said is still on the desk. Across days and weeks it resets, because nothing carries over unless the app has built something extra to carry it.

This also explains two frustrations that come up again and again in AI companion communities:

• Sudden resets. If an app stores little outside the current conversation, a long gap or a fresh session can wipe the desk clean. The companion that knew your children's names on Tuesday has no idea who you are on Friday.
• Personality drift after updates. When an app moves to a new underlying model, the same stored words get read by a different brain. Without a strong memory layer underneath, nothing anchors the companion to your shared history, so the change can feel total.

The answer is not a bigger desk, although bigger context windows help at the edges. The answer is a memory system that lives outside the conversation.

How memory works inside a companion app: what we built

This part is first-hand. We designed and built the memory system in My Softly ourselves, and what follows is how it actually works, described in plain words. The same four layers are worth looking for in any app, whatever names it gives them.

One thing first. Memory is not written while you are typing. After you chat, a separate process reads the conversation back, with each line stamped with the date and time you sent it, and decides what is worth keeping. That is why the companion can reply quickly and still build up a careful record of you in the background.

1. A profile of your life

The first layer is a short profile of the lasting facts about you: who you are, your health as you have described it, what you like and dislike, and the people in your life, each with their name. We built it so that each subject has exactly one entry, and when something changes, that entry is rewritten rather than a new note being added beside the old one. If you say your daughter has moved to Leeds, the entry about her is updated, and there is still only one version of the truth. This profile is placed in front of the companion every time you talk.

The companion keeps a short record of its own life story in the same place, so the things it has told you about itself stay consistent from week to week.

2. Themes about how you are

Facts alone make a companion sound like it has read your file. The profile can hold "sister: Claire, had an operation in March", but it has nowhere to put "goes quiet when things get busy, and likes being asked a second time". That second kind of knowing is a theme: an observation about how you are, not a fact about what you are.

We keep these as a separate layer, one entry per theme, and each one is rewritten as the picture develops. When the set grows large, a tidy-up pass merges themes that say the same thing twice, and our code checks that nothing distinctive is lost in the merge before it is accepted. A memory is never cleared away just because it was a single moment or a single evening. This is the layer that makes a companion feel like it knows you, rather than knows about you.

3. A dated diary of your days together

For each day you talk, the app writes a short private summary of that day: what you talked about, how you were, what was left open. It is dated by your own local calendar, so a late-night conversation belongs to the right day. Once written, a diary entry is never rewritten, because it is a record of when things happened.

That matters more than it sounds. The profile holds only what is true now, but the diary is allowed to hold things that later changed: it can say you were dreading the move in April and glad of it by June, and both entries stay true. Every conversation opens with the three most recent entries in view, labelled in plain terms like "3 days ago", plus up to three older entries that relate to what you are talking about now. This is what lets Friday's conversation carry on from Tuesday's.

We also made sure memory stores real dates, never words like "tomorrow" or "next week". A note that says "tomorrow's appointment" is wrong the day after it is written. A note that says "appointment on Thursday 12 March" stays right.

4. A searchable archive of everything you have said

The fourth layer is every meaningful message you have sent, stored so it can be searched by meaning rather than by exact words. The app also keeps track of the people, places and topics you mention, so one mention can lead to the conversations connected to it.

Here is what that looks like in practice. One evening you mention that your dog Milo is off his food and you are a bit worried. A week later, you say you had a stressful morning. The archive can bring back the earlier conversation about Milo, and the companion can ask how he is doing, even though you never used the same words this week and never asked it to remember.

No single layer is enough on its own. A profile without a diary feels like a database. A diary without the archive misses the small things. Themes without facts drift into vague warmth. Together they start to resemble how a friend remembers you: the fixed facts, the sense of how you are, the running story, and being reminded of something old when it matters again.

Four details that separate real memory from marketing memory

Almost every companion app now says it has long-term memory, so the claim on its own tells you very little. Having built one of these systems, we can tell you where the real work is. It is in unglamorous details, and you can spot most of them from the outside.

• Corrections should replace, not pile up. When you say "actually, my mum's name is Carol, not Karen", a well-built system rewrites the stored entry, so there is one version of the truth. In My Softly, a correction like that is picked up in the same conversation and fixed straight away, not days later. You can check any app for this: correct a detail, wait a week, and see which version comes back.
• Memory needs to know who said what. In a long conversation two voices are talking, and only one of them is you. If the companion says "I love hiking" and a careless memory system files that under you, your profile now holds someone else's personality. We built ours so every fact saved to your profile carries the exact words it came from, and those words are checked against who said them. If they came from the companion, the fact is not saved as yours.
• Old truths belong in the diary, not the profile. You used to work at the bakery; now you don't. The dated diary should keep the bakery, because that is your history together, while the profile should hold only what is true today. Apps that mix the two either lose your history or keep bringing up your old job as if it were current.
• Storing is not the same as caring. The point of all this is not a companion that can answer "what's my dog called?" like a quiz. It is one that asks how the vet visit went without being prompted. When you compare apps, weigh what they do with memory on their own more than what they can recite when asked. We wrote separately about AI companions that text you first.

Nomi's Mind Map, explained

Nomi is the app people most often mention when memory comes up, and its Mind Map is the most visible memory feature in the category, so it deserves its own section. Everything here is from Nomi's own announcement and homepage, as of September 2026.

• What it is. Nomi describes the Mind Map as a view of your Nomi's "big-picture memory": the people, places, topics and goals that come up in your conversations, drawn as connected points. It sits on top of Nomi's short, medium and long-term memory and helps tie the details together.
• How you use it. You can explore it as an interactive graph or as a searchable table, and you can edit entries or add new ones yourself, for example to correct something your Nomi has got wrong.
• When it appears. Nomi's October 2025 announcement of Mind Map 2.0 says a Mind Map starts to show after roughly 500 messages and keeps updating as you talk. Group chats can have one too.

The Mind Map is a good example of one design choice: letting you see and edit what the companion knows. We went the other way on purpose. My Softly asks nothing of you beyond talking; you correct it the way you would correct a friend, in conversation, and it fixes the memory itself. Which suits you is a matter of taste. For a fuller side-by-side, see My Softly vs Nomi.

Test it yourself

You do not have to take any app's word for it, ours included. Memory is one of the easiest claims to check yourself. Our 21-day memory test has the full method; here is the short version.

1. Mention a few specific facts on day one. Say them in passing, not as a list. For example: your niece Freya is learning the cello, you are repainting the hallway a green you are unsure about, and you have a dentist appointment a week on Thursday. Make them specific enough that a general reply cannot fake knowing them.
2. Talk about other things for a few days. This pushes those details out of the current conversation, so anything the companion recalls later has to come from real stored memory.
3. Come at it sideways. Do not ask "do you remember what I told you?", because that invites a vague, confident-sounding guess. Try "I heard some cello music today and thought of my niece." A companion with real memory will connect it to Freya by name.
4. Watch for what it raises on its own. This is the strongest sign. Does it ever ask, unprompted, how the hallway turned out, or wish you luck before the dentist?

How to help the app you already use

If you already have a companion you like, you can often help its memory along rather than starting over. App features change often, so check each app's current help pages, but these approaches are a good start:

• Use the memory tools it gives you. Replika has its Memory tab, where you can read and remove what it has saved. Nomi has the Mind Map. Kindroid lets you write key memories, a backstory and journal entries yourself. Character.AI lets you pin important messages to Story Memory. A fact you wrote in yourself does not depend on the app noticing it.
• Repeat what matters. Things you mention in several conversations are far more likely to stick than things said once. Add a little new detail each time so it does not feel like a recording.
• Correct wrong memories straight away. Say so plainly in the moment, and in apps with editable memory fix the stored entry too. Wrong memories build on themselves if they are left.

If you are technical, or curious enough to become technical, you can also run a language model on your own computer and build exactly the memory you want on top of it. We wrote a separate guide to running a local LLM as an AI companion.

Whatever app you choose, the principle holds: a companion is only as good as its memory of you. Check before you trust. If you are still deciding which apps to try, our comparison of the best AI companion apps covers where each one stands.

Sources

• Replika Help Center, "What does my Replika remember about me?" Checked 2026-09-24. help.replika.com
• Replika Help Center, "How does Replika's memory work?" help.replika.com
• Nomi.ai homepage, on short, medium and long-term memory. Checked 2026-09-24. nomi.ai
• Nomi.ai (2025-10-09). "Mind Map 2.0: Bringing Nomi Memory Into View." nomi.ai/updates
• Nomi.ai (2025-03-11). "Major Memory Update: Expanded Capacity, Enhanced Retention." nomi.ai/updates
• Kindroid Help Center, "Memory" (persistent, cascaded and retrievable memory; journal entries). Checked 2026-09-24. kindroid.ai/docs
• Character.AI blog (2026-05-21). "Smarter Memory for Smarter Chats" (Story Memory, Facts, memory usage view). blog.character.ai
• Pi Help Center (September 1, 2026), "Pi's Memory" (Pi now remembers across chats; Manage Memories). Read 2026-10-09. help.pi.ai
• Pi Help Center, "Why does Pi seem to forget earlier messages? What conversations does it remember?" Checked 2026-09-24. help.pi.ai