Dreamz AI Review

We read a lot of compliance pages. They are almost always the same page: a vendor asserting that it follows the law, in language chosen so that nothing can be checked.

Dreamz publishes a different kind of document. It describes, step by step, what its product does when someone in a conversation shows signs of wanting to hurt themselves, including which component writes the phone number. That one detail is the reason this review opens the way it does.

What is Dreamz AI?

Dreamz is a companion chat platform: pick or build a companion, talk, and generate images and video inside the conversation. It is built in SvelteKit, runs its API on api.dreamz.ai, and sells a monthly, quarterly or annual plan plus a gem currency for premium media.

Its shelf of pre-made companions is small. Twelve, each with a full name, a stated age between 22 and 40, and a two-line setup, sorted into "Slow-Burn" and "Fast-Burn". Several lean on step-family framing, a stepsister two floors down, a stepmother who stands a beat closer than she used to, which is this category's house style and will either suit you or not. The plans all say "unlimited AI companions" and there is a creation flow, so the twelve are starters rather than the product.

Unusually for anything we review, this is not a new operation. The domain was registered in February 2023.

The features that actually matter

The crisis protocol, which is the headline. Published in full on the compliance page, and specific enough to be held to:

Messages are screened for indications of suicidal ideation or self-harm. Detection runs on the conversation rather than on anything you tell them about yourself, and it is not opt-in. When those signs appear, the product shows 988 for the United States and findahelpline.com for everywhere else. Then the sentence that made us stop: "These numbers are fixed text written into the product. They are never generated, chosen, or reworded by the AI."

Anyone who has watched a language model confidently invent a phone number will understand why that is the right engineering decision, and it is the one almost nobody states. The rest is equally concrete. You can close the notice with one tap, but close it three times in 24 hours and it stays without a close button, with the conversation continuing underneath. Every referral is logged with a timestamp and no message content, so the frequency can be reported to regulators. And raising crisis resources "never pauses, suspends, or restricts your account", because locking someone out mid-crisis would be the worst possible response.

We cannot verify from outside that the code does all this. What we can say is that nobody else in this category has written it down, and writing it down is how it becomes checkable.

Unit pricing, printed on the card. Every gem pack shows its price per gem: $0.01, $0.0083, $0.0075, $0.006, $0.0044. The effective cost of a credit is the single most commonly hidden number in this market, and we have spent weeks reverse-engineering it out of client code elsewhere. Here it is on the pricing page, where a buyer can compare packs in two seconds.

No upload path into generation. We scanned all 186 chunks, 3.4 MB. No faceswap, no image-to-image, no reference or init image, no ControlNet, no IP-Adapter. Seven file inputs exist, so we traced each: the AI SDK's internal format for attaching media the platform generated, form-library cleanup code, and one genuine upload component used in exactly two places, the report form and the takedown form, to attach evidence. The takedown form itself is properly built, asking for a resemblance type, a sworn statement and a signature.

A legal shelf with the unglamorous pages on it. Eight documents: privacy, terms, cookies, 2257, complaints and content removal, anti-human-trafficking, and an appeals policy, plus the crisis protocol. An appeals policy is the one nobody bothers with, because it only exists to let users contest the operator's own decisions.

A robots.txt someone thought about. Explicit imagery under /images/unlocks/ is blocked from every crawler including GPTBot, with the stated reason that it should stay "out of every index and training crawl", while the rest of the site is deliberately opened to AI crawlers to win brand mentions. The file carries a ticket number and a note to tighten it if policy changes. It is a small thing that tells you an engineer, not a template, wrote it.

Pricing, without the bullshit

All figures are from the public pricing page and match two independent third-party reviews, which is better corroboration than most numbers in this category get.

The annual plan is a genuine saving rather than a trap: 44% off the monthly rate with 50% more gems, no escalating renewal, and no evidence of the referral-dependent pricing we found at a competitor this week. Gems do not auto-renew and do not expire.

Two caveats. The FAQ answers "Is billing discreet?" with "Shhhhh, we won't tell if you won't. Credit card won't show anything related to this." The checkout does display a billing descriptor before you pay, which is good practice, but the text itself loads from a translation file we could not read, so the claim is unevidenced. And payments run through Paycom and Praxis, both high-risk gateways, which is normal for the sector and worth knowing.

What users say

Thin, and partly about something else. Several results for "Dreamz" are reviews of an unrelated product called Dream Companion, which is worth knowing if you go looking. The genuine coverage is affiliate-shaped but at least consistent: the three subscription tiers and the gem packs are reported identically across two independent sites, which is why we treat those numbers as corroborated rather than merely claimed.

The free tier is where sources fall apart. One says it exists, one says none is confirmed, and one reports 100 gems and 10 messages on a test account while warning it may vary. Dreamz's own FAQ confirms free chat exists and never says where it ends.

On the ledger: no breach, no scam report, no regulatory action, and nearly four years of continuous operation, which is the longest run of any tool we have reviewed this month.

Final verdict: who is Dreamz AI for?

Get Dreamz if: you want a companion app from an operator that behaves like a company. Named entity, real address, four-year domain, unit pricing on the shelf, an appeals policy, a takedown flow built for victims, and a crisis protocol published so it can be checked. If you or someone using your account might ever have a bad night, that last item is not a footnote.

Skip it if: you want a large cast to browse, because twelve is thin, or if the step-family house style is not for you, or if discretion on your statement is a hard requirement and you need to see the descriptor before you commit.

The formula we'd pick: use the free chat to judge the writing, then take the annual at $99.99 if it lands, since it is the rare annual plan in this category that is cheaper for honest reasons.

Rating: 7.5/10. The governance other tools claim, written down where you can check it.