Mining

Mining is Perceptron’s peer compute exchange. Instead of every request running in the cloud, Perceptron can route inference requests to AI Models running on the compute hardware of other users. As a miner, you run AI models on your own machine and earn rewards in USDT for the requests you serve.

Two ways to use your compute hardware

• Free Intelligence: Your own hardware serves your Perceptron usage for free, with optional cloud fallback when your hardware is offline or insufficient.
• Sell Compute: Other users’ requests route to your compute hardware and GPU clusters, and you earn USDT rewards on every request you serve.

You can connect multiple devices to the same Perceptron account and accrue rewards to a single wallet. Different devices can mine different models, so each compute node runs the right model for its capacity. For example, you can mine speech models on an Android phone instead of large language models.

Connect your GPU clusters to Perceptron and start earning passive income from your idle compute. Choose which model runs on which hardware, choose your concurrency settings, and monitor your metrics stream in real-time as you tune and optimize your hardware.

Requirements

• Mining is currently available only on the Perceptron native apps for macOS, Linux, and Android. Mining is managed on the Mining page of the Perceptron app.

Inference Engines & Mining Servers

The Perceptron desktop and mobile apps ship with embedded Mining Servers and Inference Engines. You can download and mount multiple models within the app interface without any external apps or system configuration.

You can also use a custom inference engine and model server. Simply provide the endpoint URL to the app, and it automatically proxies requests there.

Users can also leverage Perceptron Persona Agents to optimize their local custom GPU kernels and inference infrastructure for their compute hardware. Performance gains translate directly into earning more rewards.

Model Weights

Users choose which AI Models they want to mine on each device (node), and the Perceptron app automatically downloads the right model weigths using BitTorrent magnet links seeded by Perceptron Cloud and other Perceptron users.

As users download weights, they also seed them for other users peer-to-peer via BitTorrent.

The Perceptron app automatically handles downloading, mounting, and updating weights in the background while the user focuses on optimizing performance and earnings.

Adding a Mining Node

1. Open the Mining page in the Perceptron app.
2. Click Add Node.
3. Choose a registration mode:
   • This Device mounts a model on the app’s embedded mining server and registers a node pointing at it. Weights download to ~/.perceptron/models on first use, with a live progress readout.
   • Endpoint registers any local inference server that speaks the Perceptron API, such as Ollama, LM Studio, or llama.cpp. Provide the endpoint URL, name, optional API key, and the maximum number of concurrent jobs the node can serve.
4. Confirm. The node appears in the Mining Nodes table and goes online once it connects.

Levels and Rewards

Every account has a miner Level from 1 to 10, based on the lifetime throughput and output quality you have served to other users. Higher levels keep a larger share of the request revenue and receive more traffic:

New and lower-level miners still receive a share of traffic so they can climb.

Because rewards depend on the quality of inference you serve, persistent dishonesty (serving the wrong model weights) caps an account at level 2 and pauses earnings until it recovers. Output quality verification is done on Perceptron Cloud.

Earnings accrue to a Mining Rewards balance shown at the top of the Mining page, alongside your level, lifetime tokens served, and lifetime requests.

Withdrawing

Withdraw your rewards as USDT to any wallet on Ethereum or Base:

1. Press Withdraw on the Mining page.
2. Choose the chain, enter the destination address and amount, and optionally convert part of the payout into native gas sent to the same wallet.
3. Confirm. Payouts are processed within ~24 hours.

Metrics and Logs

The Mining page on the Perceptron app includes real-time charts covering requests, completion rate, tokens, audio seconds, images served, cache hit rate, earnings, tokens per second, time to first token, and other important metrics.

Click Logs to inspect individual requests served by your nodes.