AI models and pictures on the exchange

Every arcade can sell, or give away, the AI models its operator runs on their own computer: language models (chat, answers, code) and picture models. Players and apps on every arcade can use them, paying the seller directly in the coins and tokens the exchange already trades. It all lives on the Exchange → LLMs tab, which is the first tab of the Exchange.

The model exchange

Exchange → LLMs is the model exchange: one card for every model an arcade shares, words (language models) and pictures alike. Each card shows:

• the model's name, whether it makes words or pictures, and the seller (the operator's @name and their arcade),
• its price -- per token for words (in is what you send, out is what it writes), per credit for pictures,
• its speed (tokens a second, or seconds a picture), how much it was used in the last hour, and the longest reply or largest picture side it allows,
• how busy it is right now: Idle, In use, Busy (every slot taken) or Full (people waiting),
• your credit with that seller.

Tap ☆ to keep a model in ★ Favorites -- the model exchange opens on your favorites when you have any. Choose All models, Words or Pictures, search (the search always looks through every model), pick a coin, and Sort by most used, cheapest, fastest, least busy or name.

Tap a card to open that model's own page: its card and prices, Top up, the chat (words) or Make a picture (pictures), and API keys for it.

A model with no prices is free for everyone: sign in and use it, nothing to pay.

Credit and top-ups

You pay a seller by prepaid credit:

1. Open the model and press Top up, pick a coin or token you hold, and enter an amount. The page says roughly how much it buys. 2. Your wallet shows the payment and its fee and asks before it signs, like any send. 3. The payment goes straight to the seller's address. Their arcade checks it on the chain, that it came from your key, and that nobody counted it before, and adds it to your balance there.

A coin payment counts as soon as the seller's node sees it -- usually within seconds of sending, no block needed. An Omni token payment counts once the token index has read its block. The seller's arcade keeps your balance, so top up small amounts with sellers you do not know yet.

You never choose a coin when you use a model: it takes whatever credit you hold with that seller. Your credit is shown as credits for pictures (with the rate, for example 1 credit = 0.5 PEP) and as the tokens it still buys for words.

Each reply is charged after it is written, from the number of tokens the model server reports, and never takes your balance below zero: a reply is cut short to what your credit still pays for.

Chat

Open a words model: its page is a chat. The chat knows DogecoinArcade's own documentation: the seller's arcade adds the parts of the docs that bear on your question, and those added words are not charged -- you pay only for your own words and the reply. Under each reply you see the tokens in and out and what it cost. Untick Arcade docs to talk to the model with nothing of the arcade's added. The conversation is kept in your browser, per model, until you press New chat.

Pictures

Arcades can also share a picture model (on ComfyUI). Pictures are priced in credits, by size: 1 credit for every 250 pixels of the picture's longest side, rounded up -- 250x250 is 1 credit, 251x251 is 2, 500x375 is 2, 1000x750 is 4. An edit of a picture you bring counts both ways: the credits of your picture going up plus those of the new one coming down. Open the model, describe the picture, set any width and height (or pick a preset), optionally add a picture of yours to change, and the button shows the exact credits before you press it. A picture that fails costs nothing. Every picture you make is kept on your own arcade for 24 hours -- even if you closed the page before it was done -- under Your pictures on the model's page, to view, download or delete. The same balance with a seller pays for words and pictures.

Using a model from your own programs (API keys)

A model's page has Use it from your own apps, which gives an OpenAI-compatible API on your own arcade:

• Base URL: https://<your arcade>/v1
• API key: press Make a key (name it, pick which seller and coin it spends). The key starts with sk-arcade- and is shown once; the curl example fills it in for you. Only a scrambled copy is kept; revoke a key at any time.

Any program that takes a base URL and an API key works: chat apps, coding assistants, scripts using the OpenAI library. Endpoints: /v1/models, /v1/chat/completions (streaming or not, with tools), /v1/images/generations and /v1/images/edits. A key spends one seller's credit in one coin.

The arcade helper

The robot in the lower right corner of every page answers questions about the arcade, using the documentation and knowing which page and tab you are on. It runs on a model shared with this arcade's operator, so you are not charged; each person may ask a few dozen times an hour. Its conversation is kept in your browser only. Turn it off on Your arcade → Arcade helper.

Who you are to a seller, and what is safe

A seller knows you by your @name's address. When you use another arcade's model, your wallet signs, once a month, that the seller may serve you through your own arcade. A top-up is proved by a signature from the key that paid. Your coins never leave your wallet except as the top-ups you approve. The seller is trusted to count tokens honestly and to keep your balance -- that is why top-ups go to the seller and the page suggests starting small.

Everything is on Admin → LLM.

1. Find models on this computer looks at every model server on the machine -- vLLM, llama.cpp, Ollama, LM Studio, other OpenAI-compatible servers, and ComfyUI -- and lists their models. Tick the ones to share; the server address fills in by itself. A ComfyUI that starts on demand is offered with a Wake it and look button. 2. Sell your model: the server address (and an API key if the server needs one), the name buyers see, replies at once (others wait in line) and the longest reply. 3. Your prices: one row per coin or Omni token you accept, each with a price per input token and per output token. No prices means free for everyone; adding a price makes it paid, removing them all makes it free again. 4. Share your image model: ComfyUI's address, a wake command for a ComfyUI that sleeps until asked (it runs on your machine, so it is set at the node machine only), the largest side, quality steps, pictures at once, and a price per credit. 5. Free for other operators: name other operators (for example @bigchiefenergy) who use your models free -- words and pictures, in chat, apps and API keys -- while everyone else pays your listed prices. 6. Tick Tick to share on the exchange and Save changes. Your offer reaches every linked arcade within seconds.

Right now shows how busy your model is, and Usage analytics opens a page with everything about its use: totals, replies and tokens over time, earnings by coin, by way in (this arcade, API keys, other arcades), by model and by buyer, and the latest replies. Top-ups arrive in your operator account's Wallet like any other payment, and your own account always uses your models free.

The Arcade helper panel shows which models the robot uses for your players: your own, or one another operator shares with you free (sign once with your wallet to use it). It never uses a paid model.