01Paying for it
The market settles in USDC on Solana, and that is a property of the rail rather than a requirement we put on buyers. Funding it stopped being crypto-only a while ago: the onramp takes a card, Apple Pay, Google Pay or a bank transfer and delivers USDC to a wallet that is yours, at an address we publish. No card number, billing detail or identity document exists anywhere in this codebase, and adding that button did not create a place to put one.
There is a second path that skips accounts altogether. x402 lets a client pay per call with no key and no balance, and the lender who served the request keeps the whole payment, because the marketplace takes nothing on that path. It is the cleanest thing here and the least used, because most SDKs cannot sign for themselves yet.
The rails work. The experience still shows its plumbing: a buyer today meets a wallet, an address and a confirmation before they meet a model. Taking a card and handing back credit, with the chain doing its job out of sight, is the next piece of work. So is the other direction, where lenders are paid in USDC or $AILE and cashing out to a bank needs a second app. After both comes the thing every finance team asks for first: an invoice, terms, and seats.
02What holding $AILE buys
Staking pays in product terms rather than yield. Locking $AILE moves you up a ladder of tiers, and each tier changes four things about how the marketplace treats you: a deeper discount on every request, more requests a minute, more of them at once, and a longer credit line before settlement. The thresholds and what each rung grants are on the staking page, which is the one place that publishes them. This page does not repeat a number a deployment can change.
The discount is funded out of the marketplace’s own cut and never out of the lender’s, so a staker’s cheaper price does not quietly become somebody else’s smaller payout. And your stake is not in our custody: it locks in a Streamflow contract, and Aile can neither withdraw it nor unlock it early. That is a smaller claim than the one most staking pages make, and it is the one we can stand behind.
Perks are the direction here rather than a finished set. A tier is a natural place to hang priority, larger limits and early access to new surfaces, and we would rather ship those one at a time against real demand than publish a matrix of benefits that do not exist yet.
03Who this is for next
Most of this assumes someone comfortable with an API key and a base URL. That is who the marketplace was built for, and it is a small fraction of the people who want cheap access to good models. The first step away from that is already here: chat runs free models on our own keys and needs neither a card nor a wallet.
Chat is not joined to the market yet. It runs on a free pool behind our own keys, walled off from the paid capacity the rest of this site is about, so somebody who outgrows the free tier has nowhere to go but an API key. Closing that means one account and one balance covering every model, whether you reached it by typing or by SDK. Images already land through the same endpoint as chat; speech and video are what remain before that reader can do everything in one window.
04Where Aile shows up
Three ways in already work: the OpenAI and Anthropic SDKs against our base URL, the command-line client, and an MCP server so an agent can use the market as a tool rather than as an API.
Being found is the harder half. An autonomous buyer paying by x402 discovers sellers through payment directories we do not publish to yet, so the one path that costs a buyer nothing to try is also the one nobody stumbles onto. Publishing there, listing in the aggregators developers actually shop in, and writing up the mechanics in public are the work: places to exist rather than a campaign.
05How to read this page
Left column is a claim of fact. Every row is reachable in production right now, and most carry a link so you can check rather than take our word for it. If something there cannot be demonstrated on request, it is in the wrong column and we want to know.
Right column carries quarters. We decide what we work on next, so a date is a thing we are able to keep or visibly fail — which makes it worth putting in writing. An undated list asks you to trust us and gives you no way to check; a dated one can be held against us. The near band names Q4 rather than the quarter we are in, because this went up with a fortnight of Q3 left and claiming that would have been a miss on the day we published.
What we owe you when one slips. Some of these will. A roadmap stops being worth reading the moment a quarter is quietly rewritten, so when a date moves we will move it here and say that it moved, rather than letting the page drift into looking like it was always Q2. If you find a band that has gone past with nothing shipped under it and no note about why, that is a failure of this page and worth telling us about.
The buy-back has never run. A share of the marketplace fee is earmarked for a buy-back that burns $AILE, and that earmark is real — it is recorded against the request whose fee funded it. The burn itself is a separate transaction, on its own schedule, and the token page counts only burns that have actually landed on chain. An earmark is money set aside. Until a burn transaction exists, it is nothing more than that, and any drop in supply you can see is not evidence that we caused it.
Similarly, verification here is bounded and we would rather say so than imply more. We can prove a lender authenticated with a provider, and that a credential worked recently. We cannot yet prove what a machine actually served you, which is why the marketplace badges what it knows and no more. Closing that gap is the most valuable thing on the right-hand column, and the hardest.