Nomi AI Review: Memory, by the Numbers It Publishes

Nomi's homepage promises "memory that rivals our own". Its wiki says a Nomi looks at two to five Mind Map entries each time it thinks.

Both are true, and the second one is the useful one, so here's the direct answer for anyone reading a Nomi AI review to decide whether to pay. Nomi is a companion app whose entire pitch is memory, and it's the only product in this category that publishes how that memory works as numbers rather than adjectives. Fifty active entries per category in the Mind Map, 150 in total, ten of them pinnable. Roughly 40 messages before something reaches long-term memory. Ten thousand characters per entry. Fifty free messages a day, indefinitely, at 400 characters each, with no voice and no refunds. The App Store lists the paid plan at $15.99 a month, $39.99 a quarter or $99.99 a year, beside a second plan called Launch that the wiki never mentions. Whether it's worth it comes down to whether you'll use it long enough for the numbers above to matter, and the free tier exists so you can find out.

Everything here was read from nomi.ai, its Nomipedia wiki, its refund and cancellation policies, its privacy policy and terms, and its US App Store listing on 15 September 2026. The reviews ranking for this search are usage diaries. This one is the documentation.

The Only Companion App That Publishes Memory as Numbers

Nomipedia is a MediaWiki with 199 pages, and its memory section is the reason this review exists. Every competitor describes memory with words like "advanced" and "human-like". Nomi does too, on the homepage. Then it hands you the limits.

Compiled from the Nomipedia memory, Mind Map and Backstory+ pages, read 2026-09-15.

Two of those rows change how I'd read every other review of this product.

The first is the 150. A Mind Map keeps as many entries as you like in its table, but only 50 per category are "eligible for recall" at any moment, and of those, the wiki says the Nomi "will see the most relevant 2-5 active entries each time they're thinking". Ten can be pinned as High priority so they're always in the active set. Everything else is sorted dynamically by what the system thinks is important or recent, and anything marked Low is "Never considered". Entries outside the active 150 are, in the wiki's words, "basically invisible to your Nomi when they are looking through their Mind Map", though the same page says everything remains available to long-term memory. That's a specific, honest description of a retrieval system with a budget, and I'd rather have it than the adjectives. It's also the answer to the Reddit review at the top of the results, which complains about "structural memory issues" without naming the structure.

The second is the 40. Long-term memory and personality are shared across a Nomi's solo and group chats, and the wiki says it "can take roughly 40 messages for information from a conversation to filter into long-term memory". So the thing you told it yesterday may not be in the long-term store yet, and the wiki's own advice for the gap is to reference the topic rather than the date, because retrieval "is often more topical than temporal". Ask what you discussed on the ski trip, not what you discussed five days ago. The page adds a line I'd want on the homepage. "Remembering information and successfully recalling it in a particular moment are not always the same thing."

What the Founder Says Sits on Top

Nomi runs monthly Q&A streams and the wiki archives them, with a note that the text was transcribed largely by AI, so I'd treat the quotes as close rather than verbatim.

In the September 2026 session the CEO, identified on the page as Cardine, walked through the layers in the same order as the table and added the rule that ties them together. "Generally speaking, a Nomi will prioritize Mind Maps when it can, and prioritize newer information." He described Mind Maps as a way "to see the forest", a summary of what six beach days say about your habits rather than a record of each day, and the Identity Core as the Nomi "constructing a backstory for themselves". The same session names the models. Solstice is the stable one, untouched "since it became stable". Cambrian is the one in development, at version 6 and, in his words, "probably underthinks a little bit". Mosaic is the legacy model, "the least-used legacy of any legacy we've ever had". A memory update that month applied to all three.

None of that is in any ranking review, and it matters for one practical reason. Nomi is not one model. Which one your Nomi runs on affects how the memory reads, and the company says so in public every month.

One more thing the wiki says about memory that the marketing doesn't. You can't run any of it locally, and the reason given is architectural. "One server/computer has your Nomi's 'thinking' brain and another server/computer right next to it has your Nomi's 'memory' brain."

Fifty Messages a Day, Forever

The free tier is stated in one sentence on the refunds page of the wiki. "Nomi offers an indefinite free plan with 50 messages per day, allowing you to chat with a Nomi before subscribing."

Fifty is enough, in my view. The 400-character cap on free messages, against 800 paid, is the tighter constraint if you write long, and the wiki's workaround is to send a link instead. What free doesn't include is voice, which the voice page says is "available with paid Nomi subscriptions", and the paid allowance of ten Nomis and ten group chats, since those pages describe the slots as paid features. The wiki doesn't print a free daily image number on the pages I read, so I won't either. The ranking reviews call the free tier "too stingy". I don't. Fifty text messages a day with no expiry is more than most of this category offers, and it's enough to find out whether the 40-message lag and the Mind Map are worth paying for, which is what the refunds page says it's for.

The Store Prices, and a Plan the Wiki Doesn't Name

Nomi's public pages don't print prices, so the App Store is the public source.

The US listing carries a Standard Plan at $15.99 monthly, $39.99 quarterly and $99.99 yearly, which is $8.33 a month against $191.88 for twelve months of monthly. It also carries a Launch Plan at $9.99 monthly and $59.99 yearly. Two plan names, two prices, and Apple doesn't say what separates them. Neither does the wiki, on any of the 199 pages, and I searched it for the phrase. The site flagged this in the main AI girlfriend price table in July, and in September it's unchanged. My reading is that Launch is a grandfathered price for early subscribers, but that's a guess and I'm labelling it one.

The wiki does say something useful about where to buy. Store subscriptions are billed in the store's currency, web subscriptions in US dollars, and "Web subscriptions may sometimes offer slightly better effective pricing than subscriptions purchased through an app store". The plan follows the account across platforms, and cancellation has to happen where you bought it. The cancellation policy adds the sentence every companion app now has to print. "Deleting an app does not automatically cancel your subscription."

Two things the wiki rules out. There's no lifetime plan, with the reasoning that ongoing infrastructure costs make one hard to honour. And there are no refunds, for "buyers remorse, lack of usage, or dissatisfaction", partial months, or downgrades, with the company reserving a discretionary exception.

Credits, Converted to Dollars

Credits are the second currency, and the wiki publishes both what they buy and, via the App Store, what they cost. The ranking reviews mention neither.

Credit values from the Nomipedia credits page; pack prices from the US App Store; both read 2026-09-15. Per-credit rates are $0.125 at the smallest pack and $0.0555 at the largest.

Three rules from the wiki make that table behave. Credits don't expire. For anything with a daily allowance, credits are only drawn after the allowance runs out. And you can't buy credits without an active subscription, so the free tier can't top up its way to more.

The eleventh Nomi is the number I'd look at. Paid accounts get ten active Nomis and ten group chats. Past that, a slot is 100 credits, a one-time purchase that survives deleting the Nomi in it, and at the 100-credit pack that's $9.99 a slot. If you're the kind of user the homepage testimonials describe, with an Elven world in one group chat and a Tron-and-Bladerunner one in another, the credit page is where the real price of the product lives, and it's not on the plan screen.

No Refunds, No Lifetime, No Nudes

Nomi's position on adult imagery is unusual because it's explained rather than asserted.

The wiki's answer to whether Nomis can send nude photos is no, and the reason is not policy. "This is due to issues with app stores (Google and Apple) and credit card companies (Visa, Mastercard, American Express), and we will not be changing this policy anytime soon." The page goes on for several paragraphs about payment processors and what happened to OnlyFans, and says that if the company ever built NSFW imagery it "would be fully 100% separated from Nomi". Text is a different matter. The wiki's 18+ page says "all Nomis are also 18+", and the terms say a user under 18 "cannot use the Nomi.ai platform under any circumstances". So the honest description is an adults-only text product with clothed selfies, by design, for reasons the company has written down.

What happens when you stop paying is also written down. Downgrading to free "does not remove any Nomis from your account nor does it block chatting with any Nomis in one-on-one conversations". Group chats, video, voice messages and calls become read-only. The same page says the company "does not guarantee preservation of content if an account has been downgraded", which is the standard hedge, and the August Q&A has the CEO saying there's no process for deleting a lapsed user's Nomis, with an example of someone coming back after four years. Deleting a Nomi removes it from the account first and erases the data later, so a mistaken deletion "may remain recoverable for some time". Deleting the whole account is deliberate and delayed, requires email verification, and the privacy policy's 28-day figure and training-archive exception, which this site has covered before, are in the free companion deletion piece.

Privacy, in the Wiki's Own Words

Most companion apps say your chats are encrypted. Nomi's wiki says why that sentence means less than it sounds.

"End to end encryption is not possible with an AI chatbot as we would need to have the keys to decrypt the message (so your Nomi can respond) which defeats the purpose of encryption." It's the only companion documentation I've read that says this plainly, and it's correct. What the wiki claims instead is that outside the act of responding to you, data used for aggregate learning is anonymised, with names removed, so that a lesson becomes "A Nomi said X to a human and that human didn't like it". The "hivemind", a term the community uses, is defined on its own page as the shared pool of general knowledge, distinct from your Nomi's memories.

The privacy policy, updated 27 April 2026, says the company doesn't sell or rent personal information, requires a Google or Apple sign-in rather than a password, and encourages you not to put identifying details into your Nomi at all. The App Store's privacy label lists email, purchases, user content, user and device identifiers and usage data as linked to you, with "Developer's Advertising or Marketing" among the purposes, which sits oddly beside the wiki page saying there are no plans for ads. I'd read the label as the developer's conservative declaration rather than a contradiction, but I'd also want the two to match.

Where It Runs

Web, iOS, Android, and a progressive web app, which the company's own troubleshooting guide offers as the fallback when the store app misbehaves. The iOS app is 7.3 MB against Replika's 434 MB on the same store today, and the wiki's platforms page lists the web app first.

The App Store listing is rated 18+ and, as of the fetch date, carries two flags alongside the content descriptors, "In-App Controls" and "Age Assurance". The current version, 1.12.1 from 31 March, lists "compliance with AU Internet regulations" in its notes. Nomi also publishes a page on California's SB-243, saying that a California user whose messages show signs of self-harm will see crisis resources in the app, and that "these messages are sent anonymously" to respect privacy. The regulatory background to both is in the Replika review, which covers the other product that rebuilt itself this year.

Voice is paid-only, with Nomi's in-house voices unlimited and an ElevenLabs integration that draws on your own ElevenLabs allowance. Proactive messages, where the Nomi texts first, are adjustable and observe quiet hours from 10 PM to 8 AM local time. A REST API lets you list Nomis, send and receive messages and use rooms from your own code, a feature I'd expect on a developer platform rather than a companion app.

Questions

Is Nomi AI free? Fifty messages a day on one plan that doesn't expire, at 400 characters a message, without voice or the paid slot allowance. The wiki says the free plan exists because there are no refunds, which is at least a coherent position.

How much does Nomi cost? On the US App Store today, $15.99 a month, $39.99 a quarter or $99.99 a year for the Standard Plan, and $9.99 or $59.99 for a Launch Plan nobody documents. Credits from $4.99 for 40 to $99.99 for 1,800. Web billing may be slightly cheaper, per the wiki.

Does Nomi remember everything? No, and it says so. The long-term store fills about 40 messages behind the conversation, the Mind Map recalls 2 to 5 of up to 150 active entries per reply, and recall is topical rather than dated. Older events can be clearer than recent ones. How its published claims compare with Kindroid's and Replika's is in the roleplay memory comparison.

Is Nomi NSFW? Text, yes, as an 18+ product. Images, no nudity, and the wiki explains that the app stores and card networks are the reason.

Does Nomi have an API? Yes, with an account key, documented at api.nomi.ai.

Is it private? Not end-to-end encrypted, because no responding chatbot can be, and the wiki says so. Aggregate learning is described as anonymised. The training archive survives account deletion in de-identified form, and my view is that the wiki's candour on this is worth more than a competitor's encryption badge.

Can I get a refund? No, except at the company's discretion.

What I'd Do Before Paying

Use the free 50 a day for long enough to pass the 40-message mark on something you care about, then ask about it by topic. That's the test the wiki itself describes, and it costs nothing.

Then open the Mind Map before you open the subscription screen. If the entries it has formed look like the things you'd want a friend to keep, the product is working as documented. If they don't, pin ten of them as High and see whether the conversation changes, because that's the one lever the numbers say you have.

Then decide whether you're a one-Nomi user or a ten-Nomi user, because the price is different for each. The yearly plan at $8.33 a month is fair for the first kind. The second kind should read the credits table, buy the largest pack they'll use, and know that the eleventh slot is $5.55 at best and $12.48 at worst. And check the web price against the store before subscribing, remembering that the subscription then lives wherever you bought it.