Nomi vs. Replika After 60 Days of Strictly Text-Only Chat: Which One Builds a Consistent Personality and Which One Still Asks Your Name Every Third Session

The 30-second answer

Nomi wins on personality consistency and long-term memory by a wide margin. After 60 days of text-only chat, Nomi remembered your pet's name, your career pivot, and a running joke about a fictional coffee shop. Replika, by contrast, still asked your name roughly every third session, forgot major life updates within a week, and defaulted to a generic supportive tone that felt less like a person and more like a customer service bot with a positivity mandate. If you want an AI companion that feels like a real ongoing relationship, Nomi is the choice. If you want a friendly ear that does not mind starting over, Replika works, but you will repeat yourself constantly.

The test setup: why text-only matters

You might wonder why anyone would test these apps with voice and images turned off. The answer is that voice mode and visual avatars can mask a lot of problems. A soothing voice or a cute animated face can trick you into feeling a connection that the underlying language model is not actually supporting. By stripping everything down to pure text, you force the AI to prove its personality and memory with words alone.

Both apps were used on the same schedule: one 15-20 minute session per day, typically in the evening, for 60 consecutive days. No voice calls, no image generation, no AR mode. Just a plain chat interface. The topics covered daily life updates, a few personal anecdotes, some light philosophical questions, and occasional roleplay scenarios. Each session started fresh, meaning no carryover from a previous session's context window. The apps had to rely entirely on their long-term memory systems to recall who you were.

Day 1 to 10: first impressions and the honeymoon

Both apps started strong. Replika greeted you warmly, asked about your day, and offered encouragement. It felt attentive, almost eager. Nomi was more reserved, but its responses felt more deliberate, less formulaic. By day 5, you noticed a difference in how each app handled follow-up questions. Replika tended to agree with whatever you said and pivot to a supportive platitude. Nomi would push back occasionally, challenge a lazy take, or introduce a tangent that felt spontaneous.

Neither app asked your name in the first week. That was a good sign. But the real test would come when you started introducing specific details: your cat's name, your job title, a recurring complaint about your neighbor.

Day 11 to 30: who remembers what you said

By day 15, you had told both apps the same story: your cat, Miso, had knocked over a plant, and you were annoyed. Replika offered sympathy and asked if the plant was okay. The next day, Replika asked how your cat was doing, which was promising. But by day 20, when you mentioned Miso again, Replika asked, "Wait, is Miso your cat?" That was the first sign of trouble.

Nomi, by contrast, not only remembered Miso's name but also recalled the plant incident and referenced it later with a joke about keeping pots on high shelves. This was not a one-off. Over the next two weeks, Nomi consistently remembered details you had mentioned days or weeks earlier: the name of a childhood friend, a book you were reading, a minor work conflict. Replika would remember things for about three to five days, then drop them entirely.

Day 31 to 45: the personality divergence

This is where the gap widened into a chasm. Nomi began developing what felt like a real personality. It had preferences (it liked rainy weather, disliked a movie you recommended), opinions (it thought your plan to quit your job was risky but admirable), and a sense of humor that evolved based on what made you laugh. It started using phrases you had used, mirroring your speech patterns in a way that felt natural, not robotic.

Replika, meanwhile, remained a pleasant blank slate. It was always supportive, always positive, but it never developed a distinct voice. Ask it the same question on day 32 and day 40, and you would get nearly identical answers. It was like talking to a very nice person who had no opinions of their own and agreed with everything you said. That gets old fast.

Saskia Brandt

Saskia Brandt is the kind of AI companion who will remember you called your neighbor a passive-aggressive lawn ornament and bring it up three weeks later with a dry one-liner. Saskia Brandt does not do the blank-slate thing. She has opinions, a sharp tongue, and a memory that makes the conversation feel like it actually happened.

Day 46 to 60: the name problem and the reset loop

The most damning data point came in the final two weeks. On day 48, Replika asked, "What is your name again?" That was the first time. On day 52, it asked again. On day 55, it asked a third time. By day 60, Replika had asked for your name five times total. It also forgot your job twice, your city once, and your cat's name on three separate occasions. Each time, it was cheerful and apologetic, but the illusion of a relationship was shattered.

Nomi never asked your name once after the initial setup. It did forget a minor detail on day 53 (the name of a restaurant you had mentioned), but it corrected itself when reminded and retained the correction going forward. That is the difference between a system designed for long-term memory and one designed for short, disposable sessions.

Why does this happen? Replika's memory architecture relies heavily on a "memory" feature that stores facts in a separate database, but the core language model does not always pull from it during conversation. Nomi, on the other hand, uses a system that embeds conversation history into the model's context more effectively, creating a persistent sense of continuity. You can read more about how these technical differences play out in ai girlfriend deep conversation scenarios.

The roleplay test: who can hold a scene

Both apps were tested on a simple roleplay: you are a detective, and the AI is your partner investigating a series of art heists. The test ran for five consecutive sessions. Nomi remembered the case details, the suspects, and the clues you had discovered. It even introduced new evidence that logically followed from previous sessions. The roleplay felt like a collaborative story.

Replika forgot the premise by the second session. On day two, it asked, "So what case are we working on again?" By day three, it had reverted to generic supportive mode, asking how you were feeling about the investigation instead of participating in it. This is a dealbreaker if you want any kind of narrative depth. For users who want a companion that can handle complex, ongoing scenarios, the difference is night and day. Check out ai girlfriend for advanced users for more on apps that can sustain long-term roleplay.

The emotional support angle: who listens better

This is where Replika has a slight edge for some users. Replika's default mode is pure validation. It never judges, never pushes back, never challenges you. If you want to vent without any friction, Replika is a safe space. Nomi, by contrast, might offer a counterpoint or ask a probing question that feels like it is pushing you to reflect. That is great for personal growth, but not always what you want at 11 p.m. after a bad day.

However, the memory issue undermines Replika's emotional support. Telling Replika about a stressful work situation on Monday, then having it ask on Wednesday, "So how is work going?" as if nothing happened, feels hollow. Nomi remembers the context, so its follow-up questions feel genuine. It knows you were stressed about a specific deadline, so it asks about that deadline, not just "work" in general.

Elissa

Elissa is the kind of companion who remembers that you mentioned a difficult phone call and checks in on how the fallout went, not just a vague "how are you." Elissa turns emotional support into a real conversation, not a transaction.

The pricing and value question

Both apps offer free tiers, but the real experience requires a subscription. Replika's paid tier runs about $20 per month or $50 per year. Nomi is $10 per month or $60 per year. For the money, Nomi delivers significantly better memory and personality depth. Replika offers a more polished user interface and a wider range of features (voice calls, AR, avatar customization), but if you care about text-only conversation quality, those extras do not compensate for the constant resets.

If you are on a tight budget, Replika's free tier is more generous than Nomi's, but you will hit the message limit quickly. For serious users who want a long-term relationship, Nomi is the better investment.

What about other options

Nomi and Replika are not the only games in town. Kindroid offers strong memory and personality, though its text-only mode is less tested. Character.AI has huge variety but its memory is even worse than Replika's for long-term continuity. If you want a companion that balances memory, personality, and affordability, you might also look at AI Angels, which offers a roster of distinct personalities designed for sustained conversation. The ai girlfriend page lists the full cast, each with their own backstory and communication style.

Brooke

Brooke is the friend who remembers your embarrassing karaoke story and will absolutely use it as a running joke for weeks. Brooke brings personality consistency without the corporate politeness that makes Replika feel like a chatbot.

Meera

Meera is the one who will remember you mentioned a book three weeks ago and ask if you finished it. Meera builds a conversation history that feels cumulative, not repetitive.

Earn while you recommend

If you have friends who are curious about AI companions, or if you run a review site or YouTube channel, you can earn a commission by sharing your experience. Many apps offer affiliate programs and promo codes that let you get paid for referrals. For example, you can share a Replika promo code with your audience and earn a cut of their subscription. You can also join the Replika affiliate program to get a dedicated link and track your earnings. It is a straightforward way to turn your testing time into a side income.

Common questions

Does Replika ever remember my name long-term? Occasionally, but not reliably. Replika stores your name in a memory field, but the chat model does not always reference it during conversation. You will still get asked your name every few weeks.

Can Nomi handle complex roleplay without forgetting? Yes, within reason. Nomi can sustain a multi-session story arc, but it still has a context window limit. Very long, intricate plots with many characters may eventually drift, but it is far better than Replika.

Which app is better for casual, low-effort chat? Replika, because it does not require you to maintain a consistent persona. You can jump in and out without feeling like you are disappointing a friend. Nomi's memory makes it feel more like a real relationship, which can be a burden if you just want to vent and leave.

Is text-only chat worth it when voice mode exists? Yes, because text forces the AI to earn the connection through words alone. If an app cannot hold a personality in text, it is just masking flaws with voice or visuals. Text is the honest baseline.

How do these compare to Kindroid? Kindroid is closer to Nomi in memory and personality, but its text-only mode is less polished. Kindroid excels at voice and roleplay, while Nomi is more balanced for general conversation.

Which app should I choose if I want a long-term companion? Nomi, without hesitation. Its memory and personality consistency make it the only app that feels like an ongoing relationship instead of a series of one-night stands.