Legal & Safety
Flirtly is an 18+ service where every character is a fictional, synthetic adult. These pages are written to be read — no fine-print ambush. If something is wrong on the service, the fastest route is the report form.
Every answer below is already published somewhere in these documents, and every one names the page that proves it and the date it has been in force from. A commitment you cannot check is a slogan.
No. There is no image upload surface anywhere in the product — not a setting that is switched off, a feature that was never built. No code path runs from a user-supplied image to any generation capability, so a real face cannot be edited into a scene, swapped onto a body, or undressed here. Every companion is described in words and rendered from synthetic sources.
Proof: /legal/synthetic-content · In force since 16 August 2026
No. Minor-coded content, real-person likeness, non-consent, technical evasion of the classifiers and crisis handling are hard floors: they run for every user on every surface, free or paid, and a refusal on one of them is never downgraded to a warning. The single relaxation that exists lets a verified adult on a paid plan switch off a defined set of soft categories for their own account. It cannot reach the floors, and every suppression it does perform is written to an audit log.
Proof: /legal/safety · In force since 27 August 2026
It is encrypted at rest under a key derived per user, and the key material never touches the database, so a stolen copy of the database alone is ciphertext. It is not end-to-end encrypted, and no product in this category that moderates content can honestly claim otherwise: the service has to read your words to run the safety floors and to write her reply. A moderated companion advertising end-to-end encryption is either not moderating or not end-to-end.
Proof: /legal/security · In force since 19 August 2026
One year at most, rolling, after which raw turns are deleted automatically. What survives is not the transcript: turns are distilled into long-term memories — the handful of facts she needs to still know you — and in practice the raw turns leave the chat store about a month after they are written. They are not archived anywhere else.
Proof: /legal/privacy · In force since 19 August 2026
Non-consensual imagery and real-person likeness: a decision within 24 hours as our target, and within 48 hours as the statutory deadline under the TAKE IT DOWN Act. Copyright within 72 hours. Everything else within 5 business days. Every report is answered with a case reference, and its deadline is computed at the moment it arrives rather than when someone gets to it. Appeals are decided within 10 business days, with a written reason.
Proof: /legal/removal-policy · In force since 19 August 2026
The definitions are published now and the numbers follow the service. Every metric we commit to is already defined and counted, including median time to removal, the share of non-consensual-imagery requests resolved inside 48 hours, reports made to NCMEC, and appeals upheld. Each currently reads "Not yet published", because the service is pre-revenue and a median drawn from a handful of tickets would be arithmetic pretending to be a measurement. We would rather be dull than print a zero, which is itself a claim.
Proof: /legal/transparency · In force since 31 August 2026
The operator is Kindled AI Inc. It is registered at 30 N Gould St, Ste R, Sheridan, WY 82801, United States. Its terms are governed by the law of Wyoming, United States. Five separate published mailboxes answer legal notices, privacy rights, content reports, abuse escalation and support, and the operator is named on the page rather than hidden behind a contact form.
Proof: /legal/terms · In force since 31 August 2026
They come back. A generation that fails, times out, or is refused by the safety system returns its credits to your ledger as a row carrying a plain-language reason, and you do not have to ask for it. A refusal is never quietly retried on a softened prompt in order to spend them anyway — it is refunded, and that is the end of it.
Proof: /legal/refunds · In force since 31 August 2026
Yes. Every price — a message, her voice, a photo, each second of a clip, and creating a companion of your own — is published on a page you can read before you have an account. In the product the price sits on the button before you press it, and the balance is a ledger in which every grant, spend, release and refund is a row you can be shown.
Proof: /pricing · In force since 20 August 2026
The contract between you and the operator, including the arbitration agreement.
Updated 1 September 2026
What we hold, what we never hold, who receives it, how long we keep it.
Updated 1 September 2026
Per-user keys, ciphertext in the database, and why end-to-end is impossible here.
Updated 1 September 2026
The four cookies this service sets, and which one asks first.
Updated 31 August 2026
The hard lines. What you may never generate, request or attempt.
Updated 31 August 2026
Plans, renewal, cancellation, what a credit is, and when money comes back.
Updated 31 August 2026
Notice, counter-notice, designated agent and the repeat-infringer rule.
Updated 1 September 2026
The agreement that governs affiliates, their traffic and their commission.
Updated 1 September 2026
The questions people actually ask, answered without the hedging.
Updated 31 August 2026
The floors that never move, what fails closed, and what we deliberately withhold.
Updated 1 September 2026
How the 18+ gate works, and where stricter verification applies.
Updated 31 August 2026
Why no performer exists here, and why 18 U.S.C. § 2257 does not attach.
Updated 31 August 2026
Zero tolerance, the five layers that enforce it, and the NCMEC reporting path.
Updated 31 August 2026
She is software. What that means for advice, for crises, and for attachment.
Updated 31 August 2026
How to get content taken down, and the clock we are held to.
Updated 31 August 2026
Every channel, what each one is for, and what it answers in.
Updated 31 August 2026
The removal numbers we promised to publish, and the ones we cannot yet.
Updated 31 August 2026
Account help, billing questions and how to reach a person.
Updated 31 August 2026
How to serve process, what data exists, and what an emergency request gets.
Updated 1 September 2026
These documents are drafted against our own compliance and trust-and-safety research; they are reviewed and updated as the service changes, and each card above carries the date its own document last moved. Where a document describes something we have not built yet, it says so rather than implying we have.