Nothing to keep up with.
A new model lands. An endpoint retires. The argument about what an AI should be changes again. None of it reaches you. Ael picks the model, decides when to search and names the conversation — and tells you what it did either way.
No card, no account. Your conversations stay on this device until you choose otherwise.
Five decisions you were never going to enjoy making.
These are the same on Tuesday as they were in March. Three of them you cannot overrule at all, because there is nothing to overrule them with. The other two you can correct, and the page says which is which rather than rounding them all up into a slogan.
Which model answers you
Not yours to change
Eight providers sit behind one button, and the routing, the failover and the cooldowns are not settings you can reach. A better model arrives, or an endpoint retires, and the change stops at us.
When it searches the web
Not yours to change
It decides whether a question needs the live web or whether it already knows. You have never had to reach for a search toggle, and there is not one to reach for.
What it is able to do
Not yours to change
A fixed set of tools: it can read a page, look at your memory, search your history, check what it has used. We decide what a conversation is allowed to reach. You are not handed a toolbox to assemble.
Which plan fits
Yours to overrule
It measures what you actually use, works out the cheapest plan that comfortably covers you, and says so in the conversation when it is worth saying. It is also allowed to tell you to stay where you are, which on a lot of weeks is the honest answer.
You act on it or dismiss it. Ael never charges you for a plan you did not choose.
What the conversation is called
Yours to overrule
It titles every thread itself, and retitles it when the subject turns, so your sidebar is readable without you maintaining it.
Rename it yourself in the sidebar whenever you disagree.
Three things it does not get to decide.
What you ask
Nobody is deciding what you bring to it. The questions are the one input the system has no opinion about.
What it keeps
Memory is a list you can read, edit, pin and delete, on the same screen where it happened. A system that decides for you still has to show its notes.
The record of it
Chats are written to this browser first, and only mirrored to your own account storage if you sign in. Export the lot as Markdown or JSON, from the menu that deletes them.
Plans.
Ael is free while it is in development — every feature, no card, no meter to watch. Paid plans will arrive later, and existing usage habits will keep working when they do.
It accounts for itself.
The section above is only defensible because of this one. A product that makes your decisions for you has an obligation to show its work, and this is where that obligation is met.
It reports what it actually did
When Ael searches, the transcript says so in plain past tense and shows how long it took. No pulsing dots and no "thinking" theatre standing in for work that is not happening.
A stopped answer says it is partial
If you stop a reply halfway, the message tells you that what is above is a partial answer. A cut-off thought is not passed off as a finished one.
Failures name themselves
When a provider is unwell, Ael says what happened in one sentence and what to do about it in the next, rather than failing quietly or apologising at you.
Questions worth answering properly.
- Isn't deciding everything for me just a black box?
- It would be, if it did not account for itself. Every tool call lands in the transcript in plain past tense with how long it took, and the two decisions it gets wrong most often — a title and a plan — are the two you can overrule. The point is not that the machine knows better than you. It is that most of these decisions are the same every time, which makes them work rather than a choice.
- What if the thing I want is on another model?
- Then it arrives, and you never find out. That is the trade, stated plainly: you give up choosing the model and you stop having to care which one it is. If you want the model to be a decision you make, this is the wrong product, and the pricing below should not talk you out of that.
- Is my data really local?
- Mostly, and the exceptions are worth stating plainly. Conversations and memory are stored in your browser, and mirrored to your own account storage only if you sign in. Answers are generated by third-party model providers, so the text of what you send does reach them — as it would with any assistant. What never happens is your history building up in our database as a profile of you.
- Do I need an account?
- No. You can use Ael as a guest and keep everything on the device. An account adds sync between devices, and a measured allowance instead of the guest rate limit — no feature is held back from the guest path.
- What happens when I reach a limit?
- Usage is measured in rolling five-hour and seven-day windows. The meter in the app shows what you have used against both, so a limit is never a surprise — and while Ael is free in development, the limits are generous enough not to get in the way.
It is free to start.
Open it as a guest, ask something, and watch what it goes and does on its own. That is the whole pitch in one conversation.
Open Ael