Sell on OpenMayhem

Resell a model you run as a proxied seller, or serve open models natively.

View as MarkdownSuggest a change on GitHub
You need a coding agent and either a model server you may resell or a computer with a suitable graphics card.
The quickest way to start is the form on the Earn page, which fills in a prompt for your coding agent.

Resell a model you already run Proxied

  • You pay a one-time admission fee of $10 the first time you open or join a proxied market. You can pay it in TNK, USD or TAP.
  • Every later market is free for the same seller, so adding another model costs nothing.
  • Sales are settled every hour.
  1. Install. The agent installs OpenMayhem Core from GitHub. Core keeps running afterwards and stops sending you buyers while your server is down or busy.
  2. Connect. It connects to your server with your address, model name and key file, then sends one short test request.
  3. Join a market. It joins the market for your model, or opens one if nobody sells that model yet.
  4. Publish. You check the listing before it goes live. The agent then publishes it and sets up your payouts.
mayhem provider proxy setup init --name my-model \
  --tokenizer-file "$TOKENIZER_FILE" --admission-origin "$ADMISSION_ORIGIN"
mayhem provider proxy setup list
mayhem provider proxy setup wizard --name my-model

Run open models on your own hardware Native

  • Native providers never pay an admission fee.
  • You keep 85% of each sale. The network fee of 15% pays for gas and operations.
  • Sales are settled every hour.
  1. Higher trust tiers. If you can prove a unique machine (T2), confidential hardware (T3) or a verified business (T4), you sell in that tier's market, which has its own price.
  2. Your own limits. You can set a daily budget and a limit on parallel requests. Turning work down because of your own limits does not hurt your reputation.
  3. Several models on one machine. A large machine can run several models at once, and each one sells in its own market.
  4. Stopping. Draining stops new sessions and lets the running ones finish.
mayhem doctor
mayhem up --provider
mayhem provider health
Details for experts

Each setup has its own name, so one seller can sell several models. The wizard saves your progress, and you can resume it at any time. The same wizard runs in the local provider dashboard, which you start with mayhem use --proxy-setup.

Paying the fee and publishing do not set up payouts. Buyers are sent to you only after at least one of your payment options has a verified payout account.

Chat models need a local tokenizer.json that matches your model. It is used to measure speed and has nothing to do with billing, and decision models do not need one. You set a budget for test requests once, and later tests can only use what is left of it. The concurrency setting says how many requests your server can take at the same time.

You can sell OpenAI chat completions, completions and responses, and OpenMayhem decisions. Servers with their own API can use a reviewed connector recipe. Images, video and audio can only be sold natively.

For experts

PreviousAuthenticationNextEndpoint reference