"Floze, Infamous for Amnesia": How Taiwanese Small App Fuwa Squeezed Into the #40 Top-Grossing Spot With 2K Installs

Fuwa, an AI romance chat app from a small Taiwanese company, has only 2K installs on Google Play but consistently sits around #40 on Taiwan's entertainment top-grossing chart. It hasn't run ads or gotten press. It rode a timing window: the biggest local competitor was being called "amnesiac," and a group of heavy players and character creators were looking for a new home. Below: how it caught them, how it charges, who competes with it, and whether outsiders should follow.

A Threads post from October 2

In October 2, Taiwanese Threads user @wen_tsou59 posted: "Recently I quit Floze because its amnesia problem got so bad, and switched to Fuwa. I didn't expect to see so many of Floze's ㄤ also showing up on Fuwa, I really almost cried… (Could replay them without amnesia.)" "ㄤ" is Taiwanese girl-circle slang for "husband," referring to the AI characters she'd chatted with on Floze. She listed 6 names and asked who else had moved over. The post got 146 likes.

Around the same time, someone on Plurk asked: "Why does my Threads feed keep recommending Fuwa? Is it actually good or sponsored?" Replies: "The top part of my feed has been flooded with Fuwa" "Lots of posts on my feed compare Floze with Fuwa" "My Threads got flooded with Fuwa too" ("脆" is slang for Threads). Another person said they'd seen a post that simply read "Floze sucks, hope Floze's creators come over to Fuwa and create characters."

These two posts more or less trace Fuwa's growth from start to finish.

What it is

A women-oriented AI romance roleplay app: you pick a male character (the store page reads "sunny dog-type college boy," "lazy cat-type senior," "and some you really shouldn't get too close to") and go back and forth with him in text to play out a story. The official pitch is "branching world lines": after any line you can return to that node and choose again, each branch keeps its own memory, and there are multiple endings and hidden achievements.

The other half is user-made content. Players can "craft" characters, writing personas, opening lines, and worldbuilding, then publish them for others to chat with. Characters and creators both have share links, and users can follow creators.

The developer is Fuwachat Technology Co., Ltd., registered in Taiwan. The iOS version launched in Jul 2026 18, and the Google Play version in Aug 23.

How big the numbers are

Ranks only in Taiwan. Taiwanese users are mostly on iPhone, so iOS is the main market, and the 2K Play installs are only a small part of it.

Revenue isn't public. For reference, on the same Taiwan Play Entertainment Top Grossing chart, Touchie (50k installs) ranks 2.82M installs and BeMi Love ranks 31 (11-06). The bar for a top 40 spot in Taiwan entertainment isn't high. Fuwa's combined monthly revenue across both platforms is probably in the hundreds of thousands of NT$ (inferred, not verified). Revenue is modest but the spend is high because of the pay-per-message model: heavy users chat hundreds of times a day and become big spenders. One App Store reviewer wrote "I'm hooked after three days, I've spent over a thousand," and another's review title reads "First AI chat purchase, dedicated to Fuwa."

Why now

The leader stumbled. The biggest Taiwanese AI romance app for women is Floze from Japan's Printage (iOS Taiwan 3,752 ratings, 289k Play installs). In 2026 fall, players piled on Floze for "amnesia": mid-conversation, characters forgot what had just happened. For anyone writing a long romance story, that's fatal. Meanwhile CRUSH was questioned for "not being Taiwanese-owned," with a Plurk user writing "I got scammed by CRUSH before, so I'm wary, it's probably a company wearing a Taiwan skin."

Fuwa's reviews are strong exactly on this point. Scroll its App Store Taiwan reviews and the titles read "memory is great," "rare to find such good memory," "memory is really, really good":

"I've tried many, and this has the best memory so far, which makes me willing to pay to support it. It remembers the custom character names and specific events even I forgot." (5 stars, 10-12) "Memory is really the most important thing for this kind of app, and FUWA does it well! ... Character mommies, go check out FUWA!" (5 stars, 9-18) "The reply quality is excellent, no obvious model feel!" (5 stars, 8-30)

Add the company's Taiwan registration, and players call it "Taiwanese software." One Plurk user ranked what they care about: "Is the company Chinese-owned > expressive writing > character variety > memory." Fuwa wins on the first and the last.

Notably, the "branching worldlines" the team pushes hardest barely come up in player talk. One person said, "The writing doesn't feel much different between models; the special thing is that literary memory tree." People came over because "Floze got worse, and this one doesn't forget," not because they can pick story branches.

Where users come from: Threads comparison posts and character creators

Threads' #虛擬聊聊 hashtag. Almost everyone in Taiwan who plays AI character roleplay hangs out under this topic on Threads. Threads' algorithm pushes similar discussions to people with the same interests, so a wave of "Floze vs. Fuwa" posts is enough to flood many people's feeds. Players were posting long recommendation threads from the launch, like @mimimirice1216_ai's Aug 1 post "新ㄉ台灣軟體ㄍㄢ超喜歡", which got 3.2K views and listed how many characters each field in the character-creation backend allows. It was written for other character creators.

Creators as distribution nodes. This is the move Fuwa should be studied for. Creators who had popular characters on Floze moved them to Fuwa, then posted link.fuwachat.com character links on Threads and IG, bringing their own fans along. @wen_tsou59 found 6 old Floze characters on Fuwa this way. In the cold-start phase, creators supplied the content and the traffic. Fuwa then built features along that path: the new version has "follow creator" and "share creator page"; in its Halloween event, creators who changed their characters' holiday avatars earned 300 crystals and got into a limited-access zone (@fuwachat, 10-22).

The team talks to players directly on Threads. The new model Lyra launched at half price for a limited time, and recruiting beta testers gave out crystals for 100 messages. When a popular model, Nyx, was pulled because its upstream supplier stopped, the team explained publicly on Threads. A November post recruiting part-time ops staff got 435 likes.

Ads: none found. Some on Plurk suspect paid promotion, but Meta's ad library shows no ads (which doesn't prove there were none), a TikTok search for "fuwa ai" turns up nothing relevant, and nobody on X mentions it. Based on current evidence, its reach came mainly from community word of mouth.

Where the money comes from

Everything is bought with "crystals." Price per message varies by model: Eros 5 crystals, Orbit 8, Lyra 10, Zenith 20. Better models cost more, and heavy users pay extra for better writing, which sorts willingness to pay into tiers. New users get 100 crystals, plus a daily check-in bonus, so the trial barrier is low.

The most common request in reviews, from late August through October, is "please add a monthly plan": "please make a monthly card," "want a monthly or annual card," "hope for a monthly subscription, I want to chat freely." Some also ask for bank transfer or convenience-store payment. This shows a group that finds per-message charging too fast and wants a capped price, which it doesn't offer yet.

Who it's competing with

A lot. Fuwa is actually a late entrant.

Figures from the App Store and Google Play, 11-06. By scale Fuwa ranks low on this list. What it has won is a reputation for the best memory in recent months, not size.

Its weaknesses

The advantage is temporary. "Good memory" is relative to Floze's current state. If Floze fixes itself, or if Fuwa's own upstream model has problems, the advantage is gone. The latter has already happened once: the Nyx model was pulled when its upstream supply stopped, and some players stayed specifically because they "liked Nyx's replies." Which underlying model it uses, and what it costs, isn't visible from outside.

Local-brand products get things wrong too. On Oct 28, a 4-star review said: "Memory is very good and it was developed in Taiwan, which I like, but why does the character's default country say China, and why does it show 'Taiwan Province of China'?" In a community where "Chinese-backed or not" ranks first, a detail like this can do real damage once it's screenshotted and shared.

Features are still basic. Reviews are still asking for chat export, chat search, auto-reply suggestions, and editable memory trees.

If it were you

Verdict (C): There's a very good playbook to learn, but copying it isn't worth it. Taiwan's women-focused AI romance market already has about a dozen products, and Fuwa is itself a late entrant. The window it caught, "Floze has amnesia," is one that's already been taken. Based on estimated revenue, a new entrant in this market would get little.

Who it suits: Teams that already run an AI character chat product (strong memory and long context) and are looking for their next small-language market, or people already active in a country's AI roleplay scene who know a group of character creators.

Who it doesn't suit: Anyone who wants to build another app like this in Taiwan; anyone without a long-term memory approach who just wraps a model API (this category competes on whether it still remembers at message 200); anyone without sustainable model costs, who would have to pull a model when upstream cuts off.

How to do it: Copy the method, not the market. Find a local, small-language AI roleplay market with a clear leader that is getting criticized, and make the leader's most criticized point (usually memory or moderation) clearly better. From day one, give character creators tools: shareable character links, creator profile pages, follow-creator, easy character migration, event rewards paid directly to creators. For pricing, use per-message charges with model tiers, and also add the monthly card Fuwa players keep asking for, offering a capped price to people who don't want to count.

How to get users: No ads. Rely on creators and comparison posts. Find where players in that country gather (in Taiwan that's Threads; elsewhere it could be X, Discord, or LINE groups). First recruit a dozen or so popular creators from the leading product to move their characters over, so their fans follow. The team should show up in the community directly and openly explain model changes and charging rules.

When to move: Watch the leader's state, not the calendar. The window is the few months when the leader's reputation is at its worst. Once the leader fixes itself, the window closes. Fuwa went from launch to being "flooded" in about two to three months.

What would mean you read it wrong: (1) Fuwa drops out of Taiwan's top 60 grossing chart, meaning the migrated players left and creator-driven traffic didn't stick; (2) Floze fixes its memory and players on Threads move back; (3) in the market you're targeting, creators you recruit don't bring paying users, or per-message model costs exceed what players will pay.