RushChat AI Review

We spent yesterday praising fuckd.ai for shipping an actual deepfake blocklist, and criticising it for one gap: its rule was written around celebrities, which leaves the person most likely to be harmed, an ex-partner or a colleague, formally outside the policy.

RushChat closes that gap in a single sentence. Its FAQ lists what it does not condone, and the entry reads: "Real life images, including real life figures such as celebrities, politicians, or people you may personally know."

Four words that no other platform in this project has written down. And they sit in a document last revised in 2024, on a site whose terms of service is shorter than this review.

What is RushChat AI?

RushChat AI is an NSFW character chat platform: a large catalogue of user-created companions across anime, realistic and fantasy styles, a character creator, and unfiltered roleplay. It claims over 10,000 companions and advertises iOS and Android apps alongside the web product. It runs in twelve languages, which is unusual generosity at this size.

The domain situation needs a paragraph. rushchat.ai is the product. rushchatai.com serves substantially the same site on a separate domain, and each one canonicalises to itself rather than to the other, so the brand is competing with its own duplicate in search results. rushchat.com redirects somewhere else entirely, and rushchat.app is a thinner landing page. Use the .ai.

The operator is STAYONE TECHNOLOGY PTE. LTD., named in the terms. The PTE LTD suffix makes it a Singapore private limited company, which is the first Singaporean operator we have documented in this run, after six Cypriot companies and one each from Estonia, Slovakia, Romania, Switzerland and the UK. No registration number is published, so unlike yesterday's Companies House filing there is nothing here to look up.

The policy question

RushChat passes cleanly, on both the architecture and the language, and the language is the interesting half.

On architecture: we pulled all ten of its application bundles, about 3.7 MB, and found nothing. No file input, no image-accepting field, no upload endpoint, no image-to-image, no reference image, no inpainting, no identity adapter, no face swap. The keyword hits that made us look twice on the homepage were an inlined Material Icons stylesheet, where swap_horiz, cloud_upload and cyclone match on substrings of icon names. There is no route by which a photograph enters this system.

On language, two rules stand out.

The first is the real-person clause quoted above. We have now read something like forty of these. The standard construction scopes the ban to "celebrities" or to "public figures", sometimes with a "deceptive or harmful" qualifier that weakens it further. RushChat's version covers celebrities, politicians, and people you may personally know, with no qualifier. That is the formulation that actually matches where the harm happens, and it costs a platform nothing to write except the willingness to think about it.

The second is its child-safety rule, which is similarly specific about loopholes. Community guidelines section 7 prohibits creating characters who are minors and posting images of minors, "including non-real depictions with a human likeness such as art, AI-generated content, fictional characters, or dolls". Naming art, AI output, fiction and dolls individually closes the four excuses this corner of the internet reaches for. Its FAQ adds that fictional characters "should be able to be glanced at and assumed to be at least 18 years old", which is a workable test rather than a slogan.

So the rules are good. Now the problem with them.

The staleness problem

Every legal document on this site carries the same vintage. Terms of service: last updated 17 May 2024. Refund and fraud policy: 17 May 2024. The community guidelines are undated but sit in the same set.

That is over two years without revision, on a platform that markets itself as a 2026 product. In that window the EU AI Act transparency obligations came into force, the UK brought in age-verification duties, and half the platforms we have reviewed this month have published documents dated 2025 or 2026 in response. ErosMate has an AI Act disclosure. eHentai has a UK age-verification update. RushChat has a page from May 2024.

The terms are also extremely thin. Stripped of navigation, the terms of service runs to barely two thousand words, which is short for a document that is supposed to govern payments, content licensing, termination and liability on an adult platform. The refund policy is shorter still.

The effect is a strange inversion. RushChat has written the single best sentence about real people that we have found anywhere, and it has written it into a body of documentation that looks abandoned. Good rules that nobody maintains are a promise with a decay rate.

The features that actually matter

A large community catalogue. Over 10,000 companions claimed, spanning anime, realistic and fantasy. Character creation is open to users, which is how catalogues that size happen, and which is also why the child-safety rule quoted above matters more here than on a platform with a curated roster.

Memory consistency. The one substantive independent review of this platform singles out its memory implementation, reporting less personality drift over long runs than competitors. We cannot test that, and we note that it is the complaint that dominates reviews of CrushOn and Candy AI, so a platform that gets it right has something real.

Twelve locales. Arabic, German, Spanish, French, Indonesian, Italian, Japanese, Korean, Polish, Portuguese, Russian and Turkish alongside English.

Mobile apps. iOS and Android are advertised, which matters because app-store distribution imposes a second layer of content review that web-only platforms escape.

Pricing, without the bullshit

Nothing on the site, again. Prices come from the API after login, and 3.7 MB of bundles contains not one figure.

One detailed third-party review reports the following, and we pass it on labelled as single-source:

Two observations if those figures hold. A free tier of 30 messages a day is genuinely usable, well ahead of Kupid AI's fifteen messages total or FLIRTcam's six, and the same source notes the free tier is new, the platform having previously been paid only. And the jump from 100 to 2,000 images between tiers is steep enough that the middle plan is either generous or the top one is aimed at a very particular user.

What we could not find anywhere, in the code or the documents, is a token or credit currency. If that is right, RushChat is one of the minority in this category charging a flat subscription rather than metering you twice. We would want to confirm that at a checkout before relying on it.

What users say

Thin, and the one real datapoint is not flattering. A detailed independent review scores RushChat 2.3 out of 5, while praising its chat quality and value, which is an odd combination we are reporting as we found it. Domain scanners rate the site as probably legitimate. There is no Trustpilot profile and no forum trail.

No breach, no scam allegations, no billing complaints. For a platform whose refund policy describes identity verification at registration and transaction, billing disputes would be the first thing to surface.

Final verdict: who is RushChat AI for?

Get RushChat if: you want a large community catalogue with unfiltered roleplay, and you value the fact that this operator thought about the private-individual case rather than just the celebrity one. If the reported free tier is accurate, 30 messages a day costs nothing to evaluate.

Skip it if: stale governance bothers you, and it reasonably might on an adult platform, or if you want a published price. Nemora publishes everything; this publishes nothing.

The formula we would pick: the free tier for a week to test the memory claim, which is the only thing anyone praises specifically, then a single month if it holds. Check at the checkout whether a coin system appears, because its absence is the quietly best thing about this platform if it is real.

Rating: 7.0/10. The right words, in a document nobody has opened since 2024.