Every AI assistant asks for an email address before it will answer a single question. It is so consistent that it reads as a technical necessity.
It isn’t. It is a consequence of one architectural decision — running the model on a server — and when you change that decision the requirement disappears entirely.
This article sits under Private ChatGPT Alternatives.
What an account is actually for
Not paranoia: cloud AI providers have genuine operational reasons for requiring accounts.
Metering. Every request consumes GPU time that costs real money. A single conversation might cost a fraction of a cent, but at scale the bill is enormous. Providers must know who is consuming what.
Billing. Subscriptions need a customer.
Rate limiting. Shared capacity must be protected from anyone consuming it without limit.
Abuse handling. Providers need to be able to ban someone, which requires an identity to ban.
Continuity. Conversation history synced across devices requires server-side storage tied to an identity.
Every one of these is legitimate. Together they make an account genuinely unavoidable — given a server.
What an account also creates
The same mechanism that makes the service workable produces something else: a durable link between one identity and everything that identity has ever asked.
Individually, most conversations are unremarkable. Collectively, a year of them is one of the more revealing datasets a person generates — often more revealing than search history, because people explain their reasoning to an assistant rather than just typing keywords. Why Your AI Conversations Are More Sensitive Than You Think covers what that record contains.
That record can then be breached, transferred in an acquisition, produced under legal process, or reprocessed under a revised policy. None of these require any bad intent from the provider. They are ordinary things that happen to stored data over time.
The account is not incidental to this. The account is what makes it a profile rather than a pile of disconnected requests.
Removing the server removes the requirement
If the model runs on your device, walk back through the list:
- Metering — nothing to meter. The compute is yours.
- Billing — no marginal cost to bill for.
- Rate limiting — the only capacity constraint is your own battery.
- Abuse handling — nothing shared to abuse.
- Continuity — history lives on your device already.
Every justification for an account was downstream of the server. Remove it and the requirement doesn’t need to be waived as a favour — it becomes meaningless.
This is why “no account” is a reliable signal rather than a marketing choice. An app that runs inference remotely essentially cannot offer it, and an app that runs locally has no reason to demand it.
How “free” works without a catch
Reasonable scepticism: free products usually mean the user is the product. Here the economics are simply different.
Cloud AI has a marginal cost per conversation — GPU seconds, electricity, bandwidth. It never reaches zero, so it must be funded by subscriptions, advertising, or data.
On-device AI moves that cost to you, in a form you don’t notice. Your phone’s compute, its battery, its storage. The developer’s marginal cost per conversation is zero. There is no recurring bill, so there is nothing that needs funding by monetizing you.
Model weights are the other half of it. Open-weight models from Alibaba, Meta, Microsoft, Mistral AI, and DeepSeek are released under licences permitting free use and redistribution — see Open Source AI Models. No per-token licence fee to recover either.
The honest limitation: you pay in device capability. Your hardware sets which models you can run, and a phone is not a data centre. That is the real trade, and it is stated plainly rather than hidden.
The practical test
You can verify all of this yourself, without reading a policy.
- Install the app.
- Download a model.
- Turn on airplane mode.
- Keep using it.
If it works, the model is on your device and nothing is being transmitted. If it stops, it isn’t.
This is the rare privacy claim that is directly falsifiable in about thirty seconds, which is worth far more than any amount of reassuring language. AI Apps That Don’t Collect Your Data covers a fuller set of tests.
What you give up, plainly
No account also means no server-side conveniences:
- No sync. Conversations live on one device. Nothing to sync them through.
- No web version. The model runs on the phone.
- No account recovery. Lose the device without a backup and the history is gone — which is the same property that makes it private.
- Device-bound capability. Your RAM sets the ceiling.
These are genuine trade-offs, not disguised features. They are also the direct consequence of the thing that makes the privacy real: if there were a server that could restore your history, there would be a server that had your history.
What it changes in practice
The subtle effect of removing the account isn’t legal or technical — it is behavioural.
When there is no profile accumulating, people ask differently. They stop sanitizing the question. They paste the actual email rather than a paraphrase, mention the real symptom, name the real company, describe the real disagreement. The tool gets used for what they actually wanted it for.
That is the point of “no account, no subscription, no sign-up.” Not the thirty seconds of form-filling it saves — the thing it stops building.
Download Cloaked on the App Store — no email, no sign-up, no subscription, and 11 models that run entirely on your device.
Frequently asked questions
Why do AI apps require an account?
Because running models on servers costs money per request. An account links usage to a payer, enforces rate limits, and prevents abuse of shared capacity. It also creates a persistent identity that ties every conversation you have to one profile.
Is there a genuinely free AI chat app with no sign-up?
Yes — apps that run models on your device rather than on servers. Because compute happens on hardware you already own, there is no per-request cost to recover, so no account or subscription is needed. Cloaked works this way.
What is the risk of using an AI account?
An account links every conversation to a durable identity, usually an email address. That creates a profile spanning months or years, which can be breached, sold in an acquisition, subpoenaed, or used to infer a great deal about you. The conversations themselves are frequently more revealing than the individual facts in them.
How can an app be free without ads or data collection?
If inference runs on your device, the developer's marginal cost per conversation is zero — you supply the compute, the electricity, and the storage. There is no ongoing bill to fund, so no advertising or data monetization is needed to cover it.