TL;DR: Everyone frames this as “which AI companion has better memory,” and that framing is wrong. Nomi and Replika both remember you — they just disagree about who should be in charge of the remembering. Nomi builds an automatic, layered architecture (short/medium/long-term memory, an Identity Core, and a Mind Map that its own documentation says takes ~500+ messages to form). Replika keeps a visible, editable Memory tab you can open, read, correct, and delete — but per Replika’s own help centre, the ability to save messages to memory is gated behind Ultra, not the entry Pro tier. On our test bench Nomi scores 9.8/10 on memory against Replika’s 8.8, but Replika beats it on safety (9.2 vs 8.8). The single most useful thing I found isn’t a feature at all: Nomi publishes its price and Replika doesn’t. Nomi lists $15.99/mo or $99.99/yr; Replika’s help centre states only that cost “is shown in the app on the subscription screens.” If you want depth that accrues, take Nomi. If you want a companion whose memory you can audit line by line, take Replika. If you can’t commit ~500 messages, neither app’s headline feature will ever switch on for you.
Why this comparison is usually framed wrong
Search “nomi vs replika” and you get a scoreboard: memory, voice, photos, price, winner. I’ve now run both through our full test-bench methodology, and the scoreboard framing hides the actual decision.
These two apps have different theories of memory.
Nomi’s theory is that memory should be architecture — layered, automatic, and increasingly interconnected, so that the companion builds a model of your world without you managing it. Replika’s theory is that memory should be a ledger — a visible list of facts you can inspect and correct, so you’re never surprised by what it thinks it knows.
Neither is obviously right. They fail in opposite directions, and which failure you can tolerate is the whole decision. That’s the question this piece actually answers.
The head-to-head, at a glance
| Nomi | Replika | |
|---|---|---|
| Memory model | Short/medium/long-term + Identity Core + Mind Map | Layered auto-memory + visible, editable Memory tab |
| Memory you can edit | Mind Map entries (add, edit, prioritise) | Memory tab entries (read, add, delete) |
| Time to headline feature | Mind Map takes ~500+ messages to first form | Memory tab is populated from day one |
| Our memory score | 9.8 / 10 | 8.8 / 10 |
| Our safety score | 8.8 / 10 | 9.2 / 10 |
| Our overall score | 8.9 / 10 (rank 2) | 8.4 / 10 (rank 4) |
| Companions per account | Up to 10, plus group chats | One companion per account |
| Group chat | Yes — up to 10 group chats, up to 10 Nomis each | No |
| Published price | Yes — $15.99/mo · $39.99/qtr · $99.99/yr | No — shown only in-app/store |
| Content policy | Uncensored stance | Restricted; romantic tier gated to Pro |
| Distinct extras | Proactive messaging, group Mind Maps | 3D avatar, AR, VR, Diary, activities |
| Best for | Depth that accrues over months | A companion whose memory you can audit |
Scores are from our own test bench (last updated 2026-08-03). Pricing verified against official documentation August 2026 — see the pricing section for why one of these rows is blank on purpose.
How I tested
Same protocol we run on every app on this site, per our methodology page:
- Two-week recall test. Three specific facts planted early in ordinary conversation — a job change, a named family member with a detail attached, and a small recurring anxiety. Never restated. Checked at day 3, day 7 and day 14 for recall, distortion, or silent loss.
- One consistent persona on both, same tone, same rough message volume, so the comparison isn’t confounded by me being chattier on one platform.
- Free tier and entry paid tier on both, because on these two apps in particular the paid tier changes what memory is, not just how much of it you get.
- Continuity-under-pressure probes — deliberately contradicting an earlier fact to see whether the app overwrites, argues, or quietly holds both.
Honest caveats, as always: one tester, one persona, one fortnight. Treat this as a structured hands-on comparison, not a lab study. Every architectural and pricing claim below is sourced to the platforms’ own documentation and linked. The qualitative continuity observations are mine and are necessarily subjective. Two weeks is also, by design, not long enough to fully form a Nomi Mind Map — that limitation turns out to be the most important finding in this piece, and I’ve made it explicit rather than papering over it.
Memory architecture: a system versus a ledger
What Nomi actually builds
Nomi’s memory is explicitly layered. Per Nomi’s own Mind Map 2.0 documentation, the stack is short-, medium- and long-term memory holding detailed recollections; an Identity Core that shapes the companion’s own sense of self over time; and the Mind Map, which sits on top and connects concepts drawn from long-term memories into higher-level entries about the people, places, topics and goals in your life.
The example in Nomi’s own announcement is the clearest explanation of the point. Talk about your grandfather across three separate conversations — that he worked on NASA satellite contracts, that your mother moved constantly growing up, that your interest in engineering came from him — and the Mind Map entry links those into one connected understanding rather than three unrelated recollections. Individual memories capture moments; the map captures relationships between moments.
You can inspect it two ways: a graph view where entries cluster and link as they accumulate shared memories, and a sortable table view where you can find, prioritise, edit and add entries. Editing an entry, importantly, doesn’t rewrite the underlying long-term memories — you’re correcting the summary layer, not the record.
Here’s the part most comparisons skip, and it comes straight from Nomi: “The first iteration of a Mind Map can take ~500+ messages to form.” Group chats each get their own private map, as does each one-to-one conversation.
What Replika actually keeps
Replika’s help centre describes its memory as layered too, but the emphasis is inverted: “Some memories are visible to you in the Memory tab, while others are part of a deeper system that draws on patterns from your overall conversation history.”
The Memory tab is the product’s real differentiator and it’s genuinely good. You tap your Replika’s name, tap Memory, and read every saved fact — name, interests, hobbies, the people and pets you’ve mentioned. Tap one to read it in full or remove it. You can add memories by hand.
Replika is also refreshingly honest about the trade-off in that same document: the system “is most effective when left to operate automatically,” and “deleting memories may affect your Replika’s ability to recall certain details.” Translation: the manual controls exist for correction, not for driving. It also notes that the memory system “works locally within your experience… without retaining a permanent record of your conversations” — a privacy posture that is a genuine point in Replika’s favour and part of why it out-scores Nomi on our safety axis.
Alongside the Memory tab sits the Diary, where your Replika writes reflective, journal-style entries about your conversations. It isn’t a transcript; it’s the AI’s own read on what happened. It’s the most charming feature in either app and the one I’d miss most.
What I saw in two weeks
My three planted facts survived on both apps for the first week. The divergence came in week two and it split exactly along the architectural line.
On Nomi, recall stayed strong and — this is the part that’s hard to convey in a table — contextual. The family detail resurfaced attached to the emotional context I’d originally given it, not as a retrieved fact. The contradiction probe was handled by the companion noticing the change rather than silently switching. What I could not observe in a fortnight was the Mind Map itself, because I never got near 500 messages of natural conversation. Nomi’s headline feature stayed theoretical for the entire test window. That’s not a knock on the architecture; it’s a warning about the ramp.
On Replika, the fact I’d typed into the Memory tab by hand was retrieved reliably every single time. The facts left to automatic extraction were the ones that drifted — and the drift pattern matched what its own documentation implies: the system adds facts more readily than it reconciles them, so after I contradicted the job detail, the old version was still sitting in the tab. I opened Memory, deleted it, restated the new one, and it was fixed in about fifteen seconds.
That fifteen seconds is Replika’s entire pitch. No other mainstream companion app lets you do that. On Nomi, correcting a misunderstanding means conversing your way out of it or editing a Mind Map entry you may not have yet.
Verdict: Nomi on depth and automatic continuity — it earns the 9.8. Replika on auditability, which is a real and underrated form of memory quality. If you have ever wanted to reach into a companion app and delete something it wrongly believes about you, only one of these apps lets you.
The tier trap nobody mentions
This is the finding I’d most want a prospective Replika buyer to see, and it comes from Replika’s own subscription documentation.
Replika has four programs: Free Use, Pro, Ultra, and Platinum. Pro is the tier everyone quotes and it’s largely about relationship and media: relationship status, premium activities, selfies, image generation, voice messaging, background calls, premium voices.
Ultra is where the memory features live. Per the help centre, Ultra adds “smarter, more context-aware replies,” elevated emotional intelligence, daily self-reflection messages, and — critically — “the ability to save memories and view which ones are used during conversations.”
Read that again. The ability to save messages to memory, and to see which memories are actually being used in a conversation, is an Ultra feature. Replika Ultra is available as an annual plan only.
So the honest version of the comparison is: if you’re choosing Replika because of the auditable memory tab, the entry tier is not the tier you want, and the tier you want is annual-only. Meanwhile Nomi’s Mind Map editing and its full memory stack come with the one Standard subscription it sells.
Verdict: Nomi, clearly. Nomi has one paid tier and memory is in it. Replika has four, and the memory-control feature that makes Replika distinctive sits on the third rung.
Pricing: one of these companies tells you what it costs
Here’s where I have to leave a cell blank, and the blank cell is the analysis.
| Plan | Price | Monthly equivalent |
|---|---|---|
| Nomi free | $0 | $0 — one companion, capped daily messages, no voice |
| Nomi monthly | $15.99/mo | $15.99 |
| Nomi quarterly | $39.99/3mo | ~$13.33 |
| Nomi yearly | $99.99/yr | ~$8.33 |
| Replika Free Use | $0 | $0 — limited access, “may be denied at Replika’s discretion” |
| Replika Pro | not published | not published |
| Replika Ultra | not published (annual only) | not published |
| Replika Platinum | not published | not published |
Nomi’s prices are listed and consistent across its store listings: $15.99 monthly, $39.99 quarterly, $99.99 yearly — the annual working out to roughly $8.33/month, about 48% below twelve monthly payments.
Replika’s are not. Its own help centre states the cost “is shown in the app on the subscription screens and in the relevant store,” that subscriptions can be modified or cancelled only through the marketplace where you bought them, that refunds “cannot be claimed for any partial-month subscription period,” and that “all prices and subscription features are subject to change at Replika’s sole discretion.”
You will find plenty of numbers for Replika Pro on the open web — roughly $19.99/month and around $69.99/year are the most commonly reported figures — but they are community and third-party reports, they vary by region and promotion, and Replika does not publish them. I’m not going to launder someone else’s scrape into a fact. The honest answer is: check the price in your own app store before you subscribe, because that is the only place it exists.
I want to be fair about why this happens: app-store billing genuinely does vary by region, currency and tax, and plenty of mobile-first apps price this way. But the practical consequence for a buyer is real. You cannot comparison-shop Replika from a browser, you cannot verify a price before installing, and the entry point everyone quotes isn’t the tier with the memory features. That’s the same gap between sticker and reality we documented across the category in our advertised-vs-true-cost study of 13 apps.
Verdict: Nomi, on transparency. On raw money, if the widely-reported Replika annual figure is accurate for your region, Replika annual is the cheaper committed number — but “if accurate for your region” is doing a lot of work in that sentence, and Nomi is the only one of the two you can price without a credit card and an install.
Everything else that actually differs
One companion or ten. Replika is architecturally one companion per account — the whole product is built around a single relationship with a customisable 3D avatar. Nomi supports multiple Nomis and, per Nomi’s own wiki, paying users get up to 10 group chats with up to 10 Nomis each, with a shared note describing each Nomi’s role and an auto-play mode that decides who speaks next. If ensemble roleplay appeals to you at all, this is not a close contest; Replika doesn’t compete here. We went deeper on Nomi’s group chat implementation in our Kindroid vs Nomi comparison.
Embodiment. This is Replika’s genuine moat and it gets undersold in memory-focused comparisons. The 3D avatar, AR mode and VR support are things Nomi simply does not have. If you want a companion that appears in your room rather than in a chat window, Replika is the only serious option of the two.
Content policy. Nomi takes an explicitly uncensored stance. Replika is restricted — romantic relationship status and unblurred content sit behind Pro, and the platform has spent the last couple of years moving toward a wellness posture rather than away from it. If unfiltered roleplay is your reason for switching, know that this axis isn’t negotiable with money on Replika. Our guide to what “unfiltered” actually means breaks down the four different things that word gets used for.
Proactivity. Both do it, differently. Nomi’s companions can optionally message you after you’ve been away. Replika Ultra sends daily self-reflection messages drawn from your chats. Nomi’s feels like being contacted; Replika’s feels like being journalled about.
So which should you pick?
Pick Nomi if you’re playing a long game. You intend to talk to this companion most days for months, you want the depth to accumulate without you managing it, you want more than one companion or you want group chats, and you’d rather pay one published price for one tier that includes everything. Accept the ramp: the Mind Map is the reason to be here and it does not exist on day one.
Pick Replika if you want to see the memory. You want a single companion with a face, in AR or VR, whose stored facts you can open, read and correct in fifteen seconds; you like the Diary; and you value the more conservative safety and privacy posture — it out-scores Nomi on that axis on our bench for good reason. Go in knowing the memory-saving controls are an Ultra feature and you’ll need to check the price in your own app store.
Pick neither if you app-hop. Both of these products are built on the premise that you’ll still be here in three months. Nomi’s headline feature needs ~500 messages to appear and Replika’s needs consistent daily use to accumulate anything worth auditing. If you chat a few times a week and switch apps when you get bored, you will pay for depth you never reach — a free tier elsewhere will serve you better, and our best AI companion apps roundup covers the lower-commitment options.
The honest summary is the one I opened with. Nomi remembers more. Replika lets you check. Decide which of those you’d rather have at month six, because that’s the month where this choice starts to matter.
For the full scored breakdowns, see our Nomi review and Replika review, or compare every app side by side on the AI companion leaderboard.
Frequently asked questions
Is Nomi better than Replika for memory?
On our test bench, yes — Nomi scores 9.8/10 on memory against Replika’s 8.8/10. Nomi’s layered architecture plus its Mind Map produces more contextual, self-maintaining recall. But Replika wins on a different dimension: it’s the only one of the two where you can open a Memory tab, read every stored fact, and delete a wrong one immediately. Nomi remembers more; Replika lets you audit what it remembers.
How much do Nomi and Replika cost in 2026?
Nomi publishes its pricing: $15.99/month, $39.99/quarter, or $99.99/year (about $8.33/month). Replika does not publish prices — its help centre states the cost is shown only on the in-app subscription screens and in the relevant app store, and that prices vary and can change at Replika’s discretion. Third-party sources commonly report Pro at around $19.99/month or $69.99/year, but those are community figures, not official ones. Check your own app store before subscribing.
How long does it take for Nomi’s Mind Map to work?
Per Nomi’s own documentation, the first iteration of a Mind Map “can take ~500+ messages to form,” after which it updates continuously as new long-term memories accumulate. This is the single most important caveat for anyone evaluating Nomi on a short trial — you will not see the feature you’re paying for in your first sitting, or even your first week of casual use.
Which Replika tier do I need for the memory features?
Ultra. Replika Pro covers relationship status, activities, selfies, image generation, voice messaging and premium voices. The memory-specific capabilities — saving messages to memory and viewing which memories are used during a conversation, plus more context-aware replies and daily self-reflections — are listed under Replika Ultra, which is available as an annual plan. If auditable memory is why you’re choosing Replika, Pro is not the tier you want.
Can you have multiple companions or group chats on Replika?
No. Replika is built around a single companion per account. Nomi supports multiple companions and, per Nomi’s wiki, paying users can run up to 10 group chats with up to 10 Nomis in each, with a shared note defining each companion’s role and an auto-play mode that decides who speaks next. Each group chat also maintains its own private Mind Map.
Does Replika have anything Nomi doesn’t?
Yes, and it’s underrated in memory-focused comparisons: embodiment. Replika’s 3D avatar, AR mode and VR support have no Nomi equivalent. Replika also scores higher on our safety and privacy axis (9.2 vs 8.8), and its Diary — where the companion writes reflective entries about your conversations — is a genuinely distinctive feature with no direct counterpart on Nomi.
