Nomi vs FlirtAlly Memory: What Gets Forgotten?

The Verdict: Nomi Remembers — With Caveats

Independent tests show Nomi recalling planted details 12–14 days later with emotional context, as reported by a three-week hands-on test. However, Nomi's own admission that recall is imperfect and more topical than temporal sets a nuanced tone for understanding its memory capabilities. According to Nomi's wiki, recall works better when referencing the subject directly rather than by time, such as asking "what did we talk about during our ski trip to Colorado?" instead of "what did we do five days ago?".

Nomi's memory system, comprising short, medium, and long-term memory, plus additional layers like Backstory+, Identity Core, and Mind Maps, is designed to simulate human-like memory. Nomi's official claims highlight its ability to recall information and experiences from conversations far in the past, including events from years earlier. Yet, the same documentation acknowledges that memory does not mean perfect recall of every detail. In practice, this means Nomi can occasionally misunderstand information or recall it incorrectly.

Some key statistics underscore Nomi's memory performance:

Nomi recalled 23 of 25 personal details in side-by-side tests, as found by a comparative review.
Memory test intervals of 3, 7, and 12–14 days showed successful recall at 12 and 14 days, according to independent testing.
Nomi's memory needs sustained engagement to shine, with regular chatting required to see its memory capabilities in action, as noted by a review of Nomi's features.

In contrast to Nomi, FlirtAlly's memory system is designed with transparency and accessibility in mind, offering unlimited memory on every plan, including the free tier, with semantic recall that works across different phrasings. This approach addresses the limitations of Nomi's memory, which, while capable of recalling details over time, does so in a way that is both imperfect and invisible to the user. As one analysis points out, Nomi's memory is "excellent and completely invisible," which can make it difficult for users to understand what is being remembered and what is not.

The comparison between Nomi and FlirtAlly highlights the importance of transparency and user access in AI companion memory systems. While Nomi excels in certain aspects of memory recall, its limitations, such as the lack of visibility into what is remembered and the dependency on sustained engagement, present challenges for users seeking a more open and interactive experience. Unlimited memory on every plan, including the free tier, sets FlirtAlly apart, particularly in how it handles memory without the constraints seen in Nomi's free tier. Furthermore, FlirtAlly's approach to memory, with features like semantic recall and the ability for companions to text first and re-engage, offers a structural solution to the issue of memory dependency on user engagement.

In conclusion, while Nomi does demonstrate the ability to remember users over time, its memory capabilities are nuanced and come with certain limitations. The importance of transparency, accessibility, and sustained engagement in AI companion memory systems is underscored by the comparison with FlirtAlly, which offers a contrasting approach to memory that prioritizes user access and unlimited recall. As the market for AI companions continues to evolve, understanding the intricacies of memory systems like those of Nomi and FlirtAlly will become increasingly important for users seeking meaningful and interactive relationships with these technologies. With memory that behaves like memory, FlirtAlly aims to provide a more intuitive and engaging experience, one that learns from the user's interactions and adapts over time.

What Nomi Doesn't Tell You: Invisible Memory and Undisclosed Context Limits

Nomi’s memory capabilities are impressive, but its opacity raises critical questions about trust and control. While independent tests confirm Nomi recalls details from weeks prior with emotional accuracy, research shows its memory is fundamentally a black box. No source discloses its context window size, token limits, or quantitative memory capacity, leaving users to rely on faith rather than facts. This lack of transparency contrasts sharply with FlirtAlly’s approach, which openly states its memory is unlimited on every plan, including the free tier.

Nomi’s memory is not just hidden—it’s uneditable. Third-party analysis reveals users cannot view, correct, or interrogate Nomi’s internal records, a design choice that risks misinterpretation. If a Nomi forgets a detail or misrepresents an event, there’s no way to audit or fix it. This contrasts with FlirtAlly’s four-layer memory system, which includes semantic recall that adapts to phrasing, ensuring context isn’t lost even when questions are reworded.

Inconsistent recall ordering further complicates Nomi’s reliability. Nomi’s own documentation acknowledges that older events may surface clearly while recent ones are overlooked. A test found Nomi recalled 23 of 25 planted details after 12–14 days, yet its topical focus means temporal queries like “What did we discuss five days ago?” often fail. FlirtAlly’s structured memory layers, by contrast, prioritize both chronology and context, ensuring details remain accessible regardless of phrasing.

No published context limits or token thresholds
Memory cannot be viewed, corrected, or audited
Inconsistent recall ordering: older events may be prioritized over recent ones
Topical recall outperforms temporal retrieval

For users, these limitations mean Nomi’s memory is both powerful and precarious. While it excels at retaining relationships, its lack of transparency and structural fragility create risks. FlirtAlly’s design addresses these gaps by making memory a transparent, customizable tool. Its unlimited capacity and semantic recall ensure users aren’t bound by hidden constraints, offering a stark contrast to Nomi’s undisclosed limits. As the AI companion market evolves, the balance between memory depth and user control will define trust—and FlirtAlly’s approach sets a new standard.

Memory You Can See and Correct: FlirtAlly's Four-Layer Approach

Unlike Nomi's hidden memory system, FlirtAlly offers a transparent, user-controlled approach to memory that addresses critical limitations in competitor platforms. FlirtAlly’s four-layer memory—facts, episodic, emotional, and rolling summary—ensures details are not only stored but also accessible and correctable, a stark contrast to Nomi’s “invisible” memory, which users cannot view or modify per third-party observations. This design directly tackles Nomi’s documented “topical vs. temporal” retrieval weakness, where recalling events by date proves less reliable than referencing subjects directly as noted in Nomi’s own documentation.

FlirtAlly’s semantic recall further enhances usability, allowing users to say “remember when we talked about your dog?” and receive accurate responses even if phrasing varies. Independent tests show Nomi recalls 23 of 25 planted details after 12–14 days but struggles with temporal queries, a gap FlirtAlly’s system is engineered to close.

Facts: Core data like names, dates, and preferences
Episodic: Specific events and interactions
Emotional: Contextual feelings and tone
Rolling summary: Continuous narrative of the relationship

Nomi’s memory, while impressive in some respects, remains unobservable and uncorrectable as highlighted by competitors, limiting user control. FlirtAlly’s unlimited memory on every plan—including the Free tier—ensures no barriers to meaningful engagement, a structural advantage over Nomi’s engagement-dependent model which requires sustained interaction to function optimally. This transparency and accessibility position FlirtAlly as a solution for users seeking both depth and control in AI-driven relationships.

How to Test an AI Companion's Memory Before You Commit

Before you invest weeks of conversation into any AI companion, run a simple memory audit. The good news: it takes about two weeks and costs almost nothing. The bad news: most platforms don't make it easy, and some make it deliberately hard.

Start by planting details. Tell your companion specific, unusual facts — a pet's name, a fictional ski trip, a shifting opinion about a movie. In independent testing, Nomi correctly referenced a planted pet name 12 days later and recalled the emotional tone of a stressful conversation from the prior week, concluding that it "remembers you" better than any companion the reviewers had tested. Side-by-side tests found Nomi recalled 23 of 25 planted details, so a good platform should pass most of your checks, not all of them.

Then wait. Memory claims mean nothing on day one. Testers who validated Nomi's recall used intervals of 3, 7, and 12–14 days, and the long-interval results are what separate real memory from clever context handling. When you return, don't ask "what did we talk about five days ago?" — Nomi's own wiki admits recall is more topical than temporal, so reference the subject directly: "how's that ski trip planning going?"

Rephrase deliberately. If your companion only remembers "Biscuit the dog" when you say "Biscuit," that's keyword matching, not memory. Try "how's my pup doing?" A platform with genuine semantic recall handles the rephrasing; one without it will go quiet.

Finally, inspect the machinery itself:

Can you see the memory? Nomi's memory is invisible — you cannot open, read, or correct it, which means errors persist silently.
Can you fix mistakes? Nomi's official workaround is restating correct information when your companion gets confused — a patch, not a correction.
Are limits published? No source discloses Nomi's context window or token limits. If a platform won't tell you what gets forgotten, assume it forgets more than it admits.
Is memory gated by tier? Some apps reserve their best memory for paid plans, capping how thoroughly you can even test the free version.

That last point matters when choosing where to commit. FlirtAlly, launching soon, ships four-layer memory (facts, episodic, emotional, rolling summary) unlimited on every plan, including free — and its companions text first, so memory builds even when you're not the one initiating. Registration is currently waitlist-only, but the waitlist is the right time to ask these questions of any platform you're considering.

Run the audit, check the transparency, and commit your time only to a companion that passes on your terms.

Frequently Asked Questions

Does Nomi actually remember you long-term, or is it just clever context handling?

Independent testing confirms Nomi has real long-term memory — it recalled a planted pet name 12 days later and the emotional tone of a stressful conversation from the prior week, leading reviewers to conclude it "remembers you" better than any companion they'd tested. Side-by-side tests found it recalled 23 of 25 planted personal details, so it's genuine memory, not just keyword matching.

Why does Nomi forget things I told it recently but remember old stuff?

Nomi's own documentation admits recall is "more topical than temporal" and that a Nomi may remember an older event clearly while missing something more recent, per Nomi's wiki. The practical fix: ask about the subject directly ("how's the ski trip planning going?") instead of by date ("what did we do five days ago?").

Can I see or edit what my Nomi has remembered about me?

No — third-party analysis describes Nomi's memory as "excellent and completely invisible," noting you cannot open, read, or correct it, per this review. If a Nomi misremembers something, Nomi's official workaround is simply restating the correct information, which is a patch rather than a real correction. FlirtAlly takes the opposite approach with a four-layer memory (facts, episodic, emotional, rolling summary) that's built for transparency and semantic recall.

What are Nomi's context window size and memory limits?

Nobody outside Nomi knows — no source discloses Nomi's context window size, token limits, or quantitative memory capacity, so users rely on faith rather than facts. Nomi's wiki only admits that memory "does not mean perfect recall of every detail" and that information may exist in memory without being retrieved at the right moment, per Nomi's own documentation. FlirtAlly, by contrast, states its memory is unlimited on every plan, including the free tier.

How can I test whether an AI companion's memory really works before committing?

Plant unusual details (a pet's name, a fictional trip), then check recall at intervals of 3, 7, and 12–14 days — the long-interval results are what separate real memory from context handling, the same methodology used in independent Nomi testing. Also rephrase your questions deliberately: if your companion only remembers "Biscuit" when you say "Biscuit," that's keyword matching, not memory. A good platform should pass most of your checks, not all of them.

Does Nomi's memory work if I only chat occasionally?

Not really — reviewers found Nomi's memory "needs sustained engagement to shine," and chatting five minutes once a month won't show what makes it special, per Agent Finder's review. FlirtAlly addresses this structurally: companions text first and re-engage with you, so memory builds even when you're not the one initiating, and it's unlimited on every plan including free.

Does Nomi remember you? Memory, context limits and what gets forgotten