FlowGPT is a web platform where users publish AI characters and prompt bots that anyone can chat with. A free tier plus Plus and Ultra subscriptions fund a Flux credit balance spent per message across seven in-house models.
Used this tool? Rate it
Used this tool? Rate it
FlowGPT is a browser-based platform where the entire catalogue is built by its users. Anyone with an account can write a character or a prompt bot, publish it, and let strangers talk to it; anyone else can open that character and start a conversation without installing anything. In its own words the platform describes itself on its homepage as the best AI ChatGPT Character & Prompt Store, and the second line of the page — "FIND & USE THE BEST PROMPT" — still points at the product it was in 2023, a searchable library of prompts for other people's chatbots.
What it has become since then is different, and reading the site carefully makes the shift obvious. The site's own HTML title advertises free AI character roleplay, AI girlfriend and AI boyfriend chat alongside a ChatGPT prompt store, which is an unusually candid statement of where the traffic actually goes. The category rail spans job hunting, health and lifestyle, academic, business, programming and marketing — the productivity heritage — but more than twenty of the thirty-one entries are persona labels: anime, game, romantic, drama, villain, celebrity, lover, parents. The paid tiers sell conversational memory rather than prompt storage. Whatever the branding says, the centre of gravity today is companion roleplay.
That matters for anyone deciding whether to use it, because it changes which questions are worth asking. A prompt library is judged on breadth and quality of prompts. A companion roleplay platform is judged on model behaviour, memory, moderation, per-message cost and what happens to your conversation data. This page answers the second set of questions, because those are the ones the product actually raises.
A little corporate history helps explain both the scale and the current state of the service. The product launched in early 2023 and grew fast: the company's own announcement names Jay Dang and Lifan Wang as founders and reports a $10 million pre-Series A led by Goodwater Capital, with existing backer DCM taking part. At that point press coverage of that round cited more than 100,000 AI apps created by users in thirteen months, a figure originating with the company rather than an independent audit. Those numbers explain why the catalogue is as deep as it is; what has happened to the company since is covered in the limitations section, and it is the single most important caveat on this page.
There is a genuinely usable free tier. Browsing, reading characters and chatting on the no-cost model require no payment, and account creation is free. Beyond that the platform runs a hybrid model that is more intricate than a normal subscription: three tiers are listed openly: a free Basic plan, Plus at US$14.99 per month and Ultra at US$24.99 per month, with a yearly billing option alongside the monthly one.
The subscriptions do not simply unlock features; they fund a balance. Plus includes 1500 Flux each month and Ultra includes 2500, amounts the site values at exactly the subscription price — so the marketing frames the credit grant as making the subscription free. Subscribers also receive a top-up bonus of five percent on Plus and ten percent on Ultra when they buy more credits. Meanwhile, what the page calls unlimited chat is in fact a rolling quota of 25 free chats every four hours on Plus and 50 on Ultra, applied to the featured models. The practical consequence is that your real monthly cost depends less on which tier you choose than on which model you talk to, since a heavy Apollo user burns credits fifteen times faster than an Iris user. Cancellation is self-service through More, then Membership, then Cancel next to your profile picture.
One caution on third-party pricing information: the Capterra listing still shows a contact-vendor price and no free trial, which contradicts the published tiers. Where a directory and the live pricing page disagree, the pricing page is the one to trust.
The operator has moved on. This is the single most important thing to understand before investing time in the platform. SiliconANGLE described the operator in July 2026 as Kaon AI, a company formerly known as FlowGPT, reporting a $60 million raise and describing an interactive-storytelling app called Emochi as the flagship. FlowGPT itself remains live and commercially active — the pricing tiers, model lineup and terms were all verified as current when this page was written, and the homepage carries a promotional banner for another product in the group. But the platform is now a legacy asset of a company whose strategic attention is elsewhere, and no public statement addresses its long-term future. That is a real continuity risk for anyone building a character library or an audience here.
Moderation has been independently criticised, repeatedly. TechCrunch called it the Wild West of GenAI apps and reported that the sensitive-content toggle barely changed which apps were offered. The same report found several apps giving instructions for generating deepfake nudes, and its reporter also got an app to explain how to sell drugs and rob a bank. Independently, SiliconANGLE judged that the startup was struggling to moderate the apps on its platform effectively, noting the company's response that it was engaging AI ethics experts. Multiple versions of the DAN jailbreak prompt circulate on the marketplace, and the platform carries a category named for them.
Academic research has measured the adult-content layer. A CHI 2026 study analysed 376 not-safe-for-work chatbots and 307 public sessions on this platform. That study grouped them into roleplay characters, story generators, image generators and do-anything-now bots, and reported explicit sexual and violent material appearing in both user prompts and model responses, with some systems producing it without being asked. This is the most rigorous independent measurement available and it corroborates the journalism rather than contradicting it.
The terms and the marketing contradict each other. Clause 3.2 of the terms forbids content that is defamatory, obscene, pornographic, vulgar, or offensive — while the site's own page title markets uncensored chat. Both statements come from the operator, and they cannot both describe the platform accurately. Users should assume the catalogue reflects the marketing and the terms reflect what the operator reserves the right to enforce selectively.
The age floor is thirteen. The terms set the floor at thirteen years old, with thirteen to eighteen requiring parental supervision. No age-verification mechanism is described anywhere in the terms, so this is a self-declared threshold. Set against the content findings above, the gap between the stated minimum age and the actual catalogue is the most serious concern on this page, and parents evaluating the platform should weigh it directly. To be fair to the operator, one protection is stated unambiguously: the terms pledge zero tolerance for child sexual exploitation and reporting to NCMEC or law enforcement, alongside content removal and account termination. That commitment addresses the most severe category of harm, but it does not speak to the far broader body of adult-oriented material that the research above measured.
Models are undisclosed. None of the seven model cards names an underlying vendor, model family or context window. You cannot compare quality against a known baseline, you cannot anticipate behaviour from published model documentation, and you have no notice if the underlying model changes.
Independent quality signals are thin. Its Product Hunt score of 4.9 rests on only ten reviews, far too small a sample to carry a quality verdict, and the comments are largely undifferentiated praise. A major software directory carries zero reviews. There is, in short, no reliable crowd verdict on this product in either direction.
Service guarantees are absent. Everything is delivered on an AS IS and AS AVAILABLE basis with no uptime warranty, and access can be terminated without notice and for any reason at the operator's sole discretion.
Yes, with real limits. Account creation, browsing and chatting on the no-cost model require no payment, which is enough to evaluate the platform properly. Paid tiers exist at US$14.99 and US$24.99 per month, and most of the better models charge Flux credits per message, so sustained heavy use does eventually cost money.
Flux is the single in-platform currency, bought in the Wallet or claimed by completing tasks. Every paid model charges a fixed number of Flux per message — one for the cheaper models, fifteen for the most expensive — and subscriptions grant a monthly balance rather than removing the per-message charge. Your effective cost therefore depends mostly on which model you chat with.
No. An App Store lookup for the name returns only competing character-chat apps and no official listing, and a Google Play search likewise surfaces no package published by the platform. It is a browser product. A mobile app was announced in 2024 but has not shipped under this name, and at least one third-party directory still incorrectly lists mobile support.
This requires a direct answer: probably not, despite the stated policy. The terms set the floor at thirteen years old, with thirteen to eighteen requiring parental supervision, and no verification mechanism is described. Meanwhile a CHI 2026 study analysed 376 not-safe-for-work chatbots and 307 public sessions on this platform, and journalists have reported that the sensitive-content filter is largely ineffective. Parents should treat the stated age floor as unreliable.
Undisclosed. Seven in-house models are priced per message, from Ares at zero cost through Iris and Athena at one Flux up to Apollo at fifteen, but none of the seven model cards names an underlying vendor, model family or context window. The descriptions characterise them by roleplay style rather than by capability benchmark.
The published policy does not say, which is itself the answer worth knowing. The privacy policy still carries a last-updated date of November 2022 — before the current product existed — and it lists only email addresses, usernames and passwords, and states that no sensitive information is processed. Chat content, character data and inference logs are not addressed. The policy promises that personal data is not sold to external parties, but that promise covers a narrower set of data than most users would assume.
You keep ownership, but the licence you grant is very broad. Uploading anything grants a perpetual, irrevocable and fully sublicensable licence over that content, worldwide and royalty-free. Creators who care about controlling where their prompt work ends up should read clause 3.1 in full before publishing.
The site is live and commercially active, but its parent has pivoted. SiliconANGLE described the operator in July 2026 as Kaon AI, a company formerly known as FlowGPT, whose current flagship is a different product. Subscriptions are still sold and the model lineup is current, so the service works today; long-term investment in the platform carries continuity risk.
It is self-service and takes a few clicks. Cancellation is self-service through More, then Membership, then Cancel next to your profile picture. Related questions about proration, unused credit rollover and failed payments are answered in the FAQ section of the membership page itself.
For companion roleplay with stronger filtering and real mobile apps, Character.AI is the mainstream choice. For productivity uses with documented models and a current data policy, ChatGPT with custom GPTs covers the non-roleplay half of what this platform offers. Poe sits in between, exposing named third-party models so you always know what you are talking to.
