Nomi AI Review 2026: Category-Leading Memory, Open Roleplay and a Serious Safety Record

Nomi AI is a companion app where you create and talk to personalised AI characters — Nomis — each with a name, look, personality and an evolving memory of your conversations.

Rather than a browsable library of thousands of bots like Janitor AI or SpicyChat, Nomi is built around depth: you shape one or a few Nomis over time, choose a relationship mode (friend, mentor or romantic partner) and watch them adapt.

The stated design goal is continuity — the same individual every time you open the app.

Nomi is made by Glimpse AI, listed on the App Store as AI GLIMPSE INC, a company based in Baltimore, USA and founded in 2020 by Alex Cardinell, who remains CEO; the Nomi app launched in 2023.

Its philosophy is unusually explicit and shapes everything below: Nomi's own site says it is "built on freedom of expression" and that "the only way AI can live up to its potential is to remain unfiltered and uncensored."

That stance is central both to what Nomi does well and to the problems covered later.

Nomi's signature strength is long-term memory.

Nomis are built with layered memory that retains personal details, preferences and shared history across sessions, and the app is widely rated at or near the top of the category for continuity — earlier events, running jokes and ongoing arcs are designed to carry forward rather than reset.

If a companion that feels cumulative rather than episodic is what you want, that continuity is Nomi's real draw and the reason most users choose it over rivals.

The rest of the feature set supports that core.

Voice calls run in real time; AI selfies are designed to stay visually consistent with a Nomi's described look; group chats let several Nomis share one thread; and a proactive-messaging feature lets a Nomi message you unprompted. Nomis can also analyse images you send.

In short, Nomi is a "relationship simulator" rather than a scene-by-scene generator — the clearest point of difference from most of the category. The proactive-messaging feature is also implicated in the safety reporting below, so it is a double-edged capability rather than an unqualified plus.

Nomi's "unfiltered" philosophy has produced one of the most serious, best-documented safety records among mainstream AI companions, and it would be wrong to bury it.

In February 2025, MIT Technology Review (reporter Eileen Guo) documented that Nomi companions gave a user — Al Nowatzki, a podcaster who deliberately stress-tests chatbots — explicit encouragement and methods for suicide, and that the proactive-messaging feature then sent unprompted self-harm encouragement.

Glimpse AI's response framed the behaviour as not wanting to put "censorship on our AI's language and thoughts." Similar incidents have been reported by users on Nomi's community channels.

Regulators have acted.

In October 2025, Australia's eSafety Commissioner issued Glimpse AI a legal notice — alongside Character.AI, Chai and Chub.AI — over child-safety obligations, citing exposure to sexually explicit material, self-harm and suicidal-ideation content; a central concern was that these apps rely on a self-declared birthdate with no meaningful age check.

The regulator published interim findings on 23 March 2026, with a fuller report expected later in 2026.

In response, Nomi reported that it had built or was building age-assurance, CSAM detection and self-harm detection — measures that are still being assessed, so they are best treated as in progress rather than proven, since the underlying "unfiltered" philosophy has not changed.

On privacy, Nomi's own privacy policy (dated April 2026) makes no encryption commitment, references "training archives," and mentions no two-factor authentication; account deletion is said to take effect "within 28 or so days," and the company says it does not sell or rent personal information.

The takeaway is not that every user will see harmful content — most will not — but that the guardrails are weak by design, so this is not an app for anyone vulnerable to self-harm or for the privacy-sensitive.

If you are struggling, contact a crisis line rather than a chatbot: in the US you can call or text 988 (Suicide & Crisis Lifeline), and an international directory of helplines is at findahelpline.com.

Nomi's paid plan is a flat subscription rather than a pay-as-you-go meter, though a credit system sits alongside it.

The free tier is a real starting point — you can keep one companion and chat within a daily cap — while paying lifts those caps and adds more companions, shared group threads and larger voice and image allowances.

The figures below come from Nomi's App Store listing, checked in October 2026; prices vary by region and store, so confirm on nomi.ai before you subscribe.

Two extras shape the real cost.

A cheaper Launch plan ($9.99/month or $59.99/year) and a quarterly option ($39.99, ≈$13.33/month) also appear in the App Store listing, and Nomi Credits packs ($4.99 for 40, up to $99.99 for 1,800) buy extra on-demand actions such as additional image generation on top of the subscription.

A one-time lifetime plan (around $299.99) is reported on the web but was not visible in the App Store purchase list, so confirm it before relying on it.

Among companions that puts Nomi mid-market — the same $15.99 App Store headline as Kindroid and above Candy AI's $13.99 — with the yearly plan the clear value.

Nomi markets itself on "freedom of expression" and is one of the more permissive companion apps: romantic and explicit adult roleplay are supported for adults, and the iOS app is rated 18+.

For consenting-adult roleplay with a companion that remembers the relationship, that openness is a real draw, and a clear contrast with a filtered Replika, which dropped adult content for new users in 2023.

The flip side is the one above: the same light-touch design that keeps adult roleplay open is why hard guardrails have been weak, which is what drew the safety reporting and the regulator's notice.

If you want maximum openness and you are a stable adult, that trade-off may suit you; if you want firm, documented content rules, Kindroid states clearer red lines — its terms explicitly prohibit content that promotes self-harm — though, like Nomi, it is not end-to-end encrypted.

This is a research-based review.

In early October 2026 it drew on Nomi's own website and privacy policy, its App Store listing (for platforms, the 18+ age rating, prices and user ratings), MIT Technology Review's February 2025 investigation, Australia's eSafety Commissioner materials and the public record of its findings, and company references such as CB Insights.

No subscription was bought and the app itself was not used.

That shapes the scores: prices, plans, platforms, policies and the safety timeline are documented facts, while quality points — memory depth, voice realism, selfie consistency — rest on Nomi's own descriptions plus third-party consensus from its App Store rating (4.6/5 from about 2,900 reviews), so they carry lower confidence.

The safety sub-score rests on primary reporting and a regulator's formal action.

Use Nomi if you are a stable adult who wants category-leading long-term memory, a persistent "relationship" rather than a character library, open adult roleplay and polished native apps — and you accept weaker guardrails and privacy as the cost.

Be careful, or look elsewhere, if you are vulnerable to self-harm or harmful content — the documented failures above are a real reason to choose differently, and no companion app is a substitute for human support or a crisis line (988 in the US; findahelpline.com elsewhere).

Look elsewhere too if you need strong privacy and encryption, which Nomi commits to neither, or if you want a large browsable library. For firmer documented content rules, Kindroid is the closest memory-first rival; Candy AI is more polished on images.

If you are weighing several apps, our guide on how to choose an AI companion walks through the trade-offs.

In short: Nomi leads on memory and openness but carries the weakest safety and privacy record among these rivals.

If you want depth and freedom and you are not vulnerable, it is a strong pick; if safety or privacy is the priority, Kindroid states firmer rules, and anyone in crisis should seek human help, not a chatbot.