Inference & Workflow Router

Use AI.Or get paid to run it.

Simulated marketOpen playground

Simulated market

8 requesters · 10 providers

8 requesters · 10 providers Gateways create demand. Providers compete to serve it.

Illustrative market activity, not live network data. Eight requester gateways and ten compute providers participate in three example transactions. Providers compete on price, speed, model availability, and evidence. Selected providers stream tokens back and both sides co-sign the resulting receipt.

Choose AI models from the network and pay only for what you use. Or connect your own computer and earn when it runs models for someone else.

Open playgroundEarn with your machine

No high-end GPU needed. CPUs and older graphics cards can earn with smaller AI models.

Built for the open model ecosystem

Model organizations represented include OpenAI, Meta AI, Qwen, Mistral AI, DeepSeek, Google DeepMind, Z.ai, Moonshot AI, MiniMax, NVIDIA, Black Forest Labs, Stability AI, Tongyi-MAI, Tencent Hunyuan, Lightricks, Microsoft, Resemble AI, Hexgrad, ACE-Step, BAAI, Jina AI, Nomic AI, Hugging Face. Names and logos identify model creators and do not imply endorsement.

Why OpenMayhem

You set the rules.

Choose what a request may cost, how fast it should run, and what protection it needs. OpenMayhem finds a matching provider and records what happened.

YOUR LIMITS≤ $0.35 · T1+≥ 20 tok/scomparing 4 live offersnode-7f2anode-3e60node-c119node-a08dover your price capbelow your speed flooridleT1 · $0.35/1M · 28 tok/s

Set your limits

Choose your maximum price, minimum speed, and level of protection. If no provider matches, the request stops before you are charged.

$0.40$0.35$0.30your session · locked $0.35price per 1M tokens · updated hourly12:00demand19:00

Your price stays locked

Prices update hourly with supply and demand. Once your request starts, the agreed price cannot change.

session 3e60-4821receipt #4821tokens128cost$0.0000352agreed ratelockedyou signedprovider signedSETTLED

Proof for every request

Every completed request creates a signed record of its usage and cost, so you have evidence if something goes wrong.

CHOOSE YOUR PROTECTIONT1Usage recordssignedT2Verified hardwareattestedT3Confidential processingprompt privacyT4Verified provideridentity checkedT3Confidential processingprompt privacySELECTEDapplied before routing

Choose who you trust

Use community providers, require proof that a machine is real, keep prompts hidden from the operator, or use only verified businesses.

Compare protection options

Hardware checker

What can your computer run?

Choose your hardware. We’ll show you which AI models are a good fit.

Your matches

20 models should run

17 comfortably · 3 with lighter settings · 2 need a local test

The memory is plausible, but the FP8 / INT8 + offload profile needs an OpenMayhem benchmark first.

Model size
20B · 50-step reference
Model format
FP8 / INT8 + offload
Runs with
vLLM-Omni

This profile is provisional and needs a local OpenMayhem benchmark before serving.

Apache-2.0Explore this model Model details (opens in a new tab)

Want to be certain? mayhem doctor checks your exact setup before anything downloads.

Request a model on Reddit (opens in a new tab)Set up my machine

Privacy & payment

Choose what matters to you.

Choose signed records, verified hardware, confidential processing, or a verified business for each request.

Pick the protection you need

Choose for each request. More is not always better.

  • Usage recordsT1Signed records of what ran and what it cost.Optional
  • Verified hardwareT2Confirms the request runs on a verified machine.Optional
  • Confidential processingT3Keeps prompts private while they are processed. selected
  • Verified providerT4Confirms the business operating the service.Optional

Your prompt, your choice

Always encrypted

Protected while it travels.

Confidential with T3

Processed inside protected hardware.

Simple, visible fees

Card example · $1.00

85¢

to the provider

15¢

network fee

Don't take our word for it. The rules and payment code are public.

Read the rules (opens in a new tab)View the payment code (opens in a new tab)

Common questions

What you need to know.

Straight answers about paying, privacy, compatibility, and earning with your computer.

Get started

Set up with a coding agent.

There is no one-click installer yet. Choose what you want, copy the prompt, and paste it into Codex, Claude Code, or another coding agent.

Earn with your machine

The agent prepares your machine to run AI models for other people.

  1. 01

    Check your hardware

    Reads your GPU and available memory.

  2. 02

    Match AI models

    Finds models your hardware can run.

  3. 03

    Start providing

    Launches provider mode and checks it.

Prefer to install it yourself? Read the setup guide (opens in a new tab) · Questions? Ask the community (opens in a new tab) · Join Telegram (opens in a new tab)