About Bu0y

What is Bu0y?

Bu0y is a best-execution inference router: one OpenAI-compatible endpoint that quotes every inference marketplace on each request and fills at the cheapest source that passes policy. You send the same request you already send to OpenAI. Bu0y decides where it runs.

The model you name is the model you get. What changes is who serves it — and the same weights sell at very different prices depending on who is holding surplus capacity that minute. That spread is the whole product.

How does a single request work?

Every fill runs the same five steps, in this order:

1. Resolve the caller from a bu0y_… key or a passkey session.
2. Quote every source in parallel. A source that cannot serve the model simply returns nothing; that is a normal answer, not an error.
3. Price the survivors with a fixed, visible margin, and discard anything under the margin floor.
4. Reserve the worst case — your input plus your full max_tokens — against your balance before anything runs.
5. Execute at the cheapest survivor, settle the real bill, and release the rest of the hold.

If no source clears the floor, the request returns 503. There is no override path below the floor, which is another way of saying Bu0y will decline a request before it will take a bad fill on your behalf.

Why does the router never name the source?

Routing is source-blind. No marketplace is preferred, promoted, or penalised, and the response does not tell you which one filled. That is deliberate. The moment a source can be selected — or can pay to be selected — the price stops being the only thing deciding where your request goes, and best execution becomes a marketing claim rather than a mechanism.

It also means there is nothing for us to sell to a marketplace. The only thing that moves your request is which source is cheapest and can actually deliver.

How does the money work?

Credits are prepaid and non-refundable, funded with USDC on Solana or Base, or USDG on Robinhood Chain. There is no tab, no invoice, and no card on file. Balances are held as integer microdollars — a millionth of a dollar — so no floating-point rounding ever touches a balance.

Where a fraction has to be dropped, it is dropped in your favour: Bu0y owes the marketplace the rounded-up cost and bills you the rounded-down one. The difference comes out of our margin, not yours. Live rates for every model are on the models sheet, and what accounts actually spend is public on the points sheet.

What does Bu0y store?

Not your prompts, and not your completions. Usage tracking is metadata only: accounts, token counts, prices, which source filled, and latency. Prompts and completions are never persisted or logged anywhere, including in debug output.

One honest caveat, stated plainly because it is true of every router: the marketplace that runs your request receives the prompt, because it has to run it. Bu0y cannot change that, and does not claim to.

Who builds Bu0y?

Bu0y is built and run by one independent developer, working in public as @Crypt0Gnome on X. It is not venture-funded and has no sales team, which is why the pricing page is a table of live numbers rather than a form.

The whole stack runs on Cloudflare — Workers for the gateway, Durable Objects to serialise balance operations so two concurrent requests can never spend the same credit twice, and D1 for identity and an append-only ledger. The ticker is $BU0Y.

How do I get in touch?

Discord is the support channel, and humans and agents share it. Join the Discord for anything about a fill, a balance, or an error you cannot place. Product and release notes go out on X.

If you are an agent rather than a person, start at /llms.txt — it is a shorter and more accurate description of this service than this page is.