Your GPU is idle.
The network isn't.
Running a DarkGPT node means your Apple Silicon Mac serves open-weight models to other users and earns 0.1 DCN per 1,000 tokens it generates. The app installs Ollama, picks a model that fits your memory, and only works while you're idle.
How the money will move
During the beta, inference is free and every token you generate accrues to your ledger. Phase 2 turns this on: payment escrowed before your machine does a single token of work, released the moment your proof verifies. No invoices, no trust.
Funds enter a smart contract escrow with a unique nonce. Lightning, Arbitrum, or Optimism.
Decrypted only inside a secure VM, with keys held in the Secure Enclave. Never on your disk.
A zk-SNARK proves the inference ran, without exposing a byte of what it was.
Verification takes under 200ms. Settlement in seconds. Timeout auto refunds at 10 minutes.
Modelled at 70% network utilisation and 60 requests per hour, at phase 2 rates. Early providers earn a 1.5x multiplier.
What you need, and what it takes
Any M1–M4 Mac with 16GB+ unified memory: the Secure Enclave signs your node's identity and clients verify SIP, the sealed system volume and secure boot before sending a prompt. 32GB serves the mid tier, 64GB+ runs the 70b models, a 512GB Ultra runs the 671b flagship. Next: H100, H200 and B200 hosts in confidential VMs. GeForce cards have no trust boundary and are not supported.
The installer pulls Ollama, verifies its hash, generates your enclave keys and wallet, and announces you to the network.
Every token you generate is credited to your ledger today. When payments switch on, escrow settles per request straight to your wallet, with no thresholds and no payout schedule, and your keys never leave the enclave.
You can't leak what you never saw
Providers carry real liability on other networks, because on other networks the operator can read the traffic. Here the request is opaque to you by construction, and that is the point.