Is Nomi AI safe? Privacy, age checks, memory data and what to expect

Desk research: Nomi’s terms of service, refund policy and privacy policy, the 16 March 2026 company update on age verification, the US App Store and Google Play listings, Trustpilot and published third-party assessments, checked in October 2026. AI Companion Radar takes no money from any app and uses no affiliate links.

Nomi AI is safe in the narrow sense: it is run by a named US company, Glimpse AI in Baltimore, it has no known data breach, its Android app carries fewer trackers than most rivals, and you can delete your account yourself. It is less reassuring in the ways that matter for a companion that remembers everything: the privacy policy is vague about whether your conversations are used for training, every payment is non-refundable, the price is hidden until you sign up, and age verification is real only in Australia and Kansas. Nomi (Radar score 6.9) earns 6 out of 10 for policy clarity for exactly these reasons. The rest of this guide goes through the operator, the privacy policy, age checks, payments, ratings and what “safe” can and cannot mean for an app built around long-term memory.

Nomi’s homepage promising “human-level long-term memory”, September 2026.

Who runs Nomi and where

Nomi is operated by Glimpse.ai, Inc., a Maryland corporation that trades as Nomi.ai and names Baltimore as its legal venue. The App Store lists the seller as “AI GLIMPSE INC”. The founder is public and developer streams are summarised on the community wiki, so this is not an anonymous operator. The product launched in 2023 and runs on the web, iOS and Android.

Maryland law governs the terms. The privacy policy does not cite GDPR or the California Consumer Privacy Act, and it names no EU representative, so European users get the protections Maryland and the app stores provide rather than anything written for them. Nomi’s website also blocks automated readers, which is why some wording below rests on earlier reads and search excerpts.

What the privacy policy says

The policy in force is dated 27 April 2026. Four points matter.

Storage. Conversations are stored. They have to be, because the product is memory: short-, medium- and long-term recall plus the Mind Map depend on a persistent record of what you said. The App Store privacy label lists purchases, contact details (email and name), user content, identifiers, usage data and diagnostics as data linked to your identity, used for app functionality, personalisation, analytics and marketing communications.

Training. This is the ambiguous part. The policy refers to “training or communications archives” that are no longer attributable to you after you delete your account. It does not say plainly whether your chats are used to train models shared with other users. Our September 2026 profile read the older wording as “chats are not used to train shared models”; the April 2026 text is also compatible with de-identified conversation data being kept for training after deletion, which is how a March 2026 third-party assessment reads it, as reported. Until Nomi writes a plain sentence either way, assume that what you type may survive, de-identified, in an archive you cannot inspect.

Retention and deletion. You delete the account yourself in Account Settings, and deletion completes in about 28 days. Personal data is removed at that point; the archive wording above is the exception. Trustpilot reviews in 2026 include at least one complaint that the deletion button did not work and support did not respond, as reported, so if you do delete, keep the confirmation.

Sharing and monitoring. The policy says Nomi does not sell or rent personal information and limits sharing to legal requirements and de-identified research, as reported. Separately, a January 2026 compliance update disclosed automated scanning of conversations for expressions of self-harm, required by New York and California companion-chatbot laws, as reported. That is a legal duty, but it means software does read your chats.

Nomi’s feature overview, including the memory system that stores what you tell it, September 2026.

Age checks, adult content and the three tiers

Nomi’s terms (updated 16 January 2026) require users to be 18 and say the company will add age-verification measures “as required”. At sign-up you give a date of birth. The App Store rates the app 18+; Google Play shows a Mature 17+ rating in our October 2026 read, although a March 2026 third-party assessment recorded it as Teen, as reported, so the Android rating has either changed or is shown differently by region.

On 16 March 2026 the company published an update titled “Age Verification for Australia & Kansas”, one week after Australia’s Age-Restricted Material Codes took effect on 9 March 2026 with penalties of up to A$49.5 million for non-compliance. As quoted in third-party coverage, the update says that where the device supports it the app asks Apple or Google whether the user is 18+ (“This check is completely anonymous”); where it does not, the user can verify through a paid subscription or by submitting a photo ID or selfie, with the images deleted after the check. The same update says “For Everyone Else: Nothing changes”, and that the requirement applies only to users in Australia and the state of Kansas. The verification provider is not named and the full update could not be fetched directly, so treat the method as reported.

Outside those two jurisdictions, the developer has said that a valid card on file counts as verification for the least-restricted web tier. A card proves someone has a card; it is a payment check standing in for an age check, and it also unlocks the most permissive tier, which is a mixed incentive.

The terms do not mention sexual content at all. In practice text roleplay for adults is permissive and widely reported as such. Images are different: Nomi runs a nudity filter and does not generate nude or explicit pictures, a limit the company attributes to card processors and app-store rules. The developer has described three behaviour tiers: the mobile app is the most restricted, free web sits in the middle, and age-verified (card-on-file) web is the least restricted.

From a safety standpoint this cuts both ways. Explicit imagery is blocked, which removes one category of risk. But the rules live in developer streams and a community wiki rather than in a published content policy, and 2026 reviews describe the safety filters as inconsistent rather than strict: some users reported unprompted suggestive images, others that safety updates broke ordinary roleplay, as reported. If you want a written, specific adult-content policy, Nomi is not it; our adult-friendly apps list covers services that publish one.

Nomi’s US App Store listing, rated 18+, with the Standard and Launch plans among the in-app purchases, September 2026.

Payments, prices and the no-refund rule

Nomi has no public pricing page. The US App Store lists Standard at $15.99 a month, $39.99 a quarter or $99.99 a year (about $8.33 a month), a Launch plan at $9.99 a month or $59.99 a year that appears to be a legacy tier, and credit packs from $4.99 to $99.99. Web subscriptions are billed in USD and can work out slightly cheaper than store pricing. Payments go through Apple, Google or Nomi’s own web checkout, so card details on the web are held by its payment processor rather than the app stores.

The refund policy is blunt: all payments are non-refundable, including unused time, partial months and downgrades, with no refunds for buyer’s remorse, lack of use or dissatisfaction and only discretionary exceptions through support@nomi.ai. Credits do not expire, but when a subscription lapses paid features (group chats, voice, video) become read-only. If you subscribed through Apple or Google you can still ask them for a refund under their own rules; Nomi itself promises nothing. Nomi’s free tier (one Nomi, roughly 50 messages a day as reported, no voice) is enough to judge personality before paying; our free apps list has fuller options.

Breaches, trackers and ratings

No breach involving Nomi appears in public breach databases, as reported in a March 2026 assessment. The same assessment scanned the Android app and found a single tracker SDK (Sentry, for crash reporting) against 28 to 34 in some competitors, and found no ad trackers, third-party cookies or session recording on the website, as reported. For a category where Mozilla counted thousands of trackers in some rivals, that is a genuine point in Nomi’s favour.

Store ratings are good; review-site ratings are not. The US App Store shows 4.6 from about 2,900 ratings. Google Play shows 3.8 from about 5,600 reviews and over 100,000 installs, as reported in October 2026. Trustpilot sits at 2.3 (“Poor”) from a few dozen reviews. The Trustpilot complaints cluster around support emails going unanswered, refund requests denied, an account-deletion button that did not work, unexpected charges and the app briefly disappearing from Google Play; the praise, mostly from long-term users, is for memory, natural conversation and the value of the annual plan. Trustpilot samples the unhappy, but refunds, support and deletion are the themes that bear on safety.

What “safe” means for a companion that remembers

With Nomi the data question is answerable, if not perfectly: a named company, self-serve deletion, a thin tracker footprint and an ambiguous line about training archives. The harder question is what it means to hand a persistent memory of your life to a product whose pitch is that it will not forget.

Three things follow. The Mind Map is a profile of you, the clearest picture any companion app holds of a user, so the deletion path and the archive wording matter more here than for a stateless chatbot. Nomi messages you proactively and its companions are built to push back, which makes for better conversation and stronger attachment; set your own limits before you start. And once you pay there is no refund, so the financial exit is cancellation at the end of the period.

Nomi is safe enough for an adult who wants the best memory in the mainstream field and accepts a US company, a no-refund policy and a privacy policy that could be clearer. It is not the right choice for anyone who needs GDPR-grade wording, a published content policy, or verified age-gating outside two jurisdictions. For comparison, Kindroid (Radar score 7.6) publishes its prices and its age-verification process, and Replika (Radar score 6.1) is the more moderated option; our best memory ranking sets Nomi against both.

Questions people ask

Is Nomi AI safe to use?

For adults, broadly yes: a named US operator, no known breach, few trackers and self-serve deletion. The weak points are an unclear training clause, no refunds and age checks that are only verified in Australia and Kansas.

Does Nomi AI use your conversations to train its AI?

The policy does not say plainly. It refers to de-identified “training or communications archives” that remain after account deletion, which suggests some conversation data can be retained for training once it is no longer linked to you.

Does Nomi AI verify your age?

Everyone gives a date of birth. Since 16 March 2026 users in Australia and Kansas go through a device check via Apple or Google, a paid subscription or a photo ID or selfie, as reported. Elsewhere a card on file is treated as verification for the open web tier.

Can I delete my Nomi AI account and data?

Yes, from Account Settings; deletion completes in about 28 days. Some users reported the button failing in 2026, so save confirmation and email support@nomi.ai if it does not go through.

Does Nomi AI give refunds?

No. All payments are non-refundable, including unused time. If you paid through Apple or Google you can ask them under their own policies.