0xalank

Hashference as a Hermes model provider (`--provider hashference`, or Hashference in `hermes model`): open and frontier models on one balance, funded by mining, QUAI, Qi or staking. Each model's real limits come from Hashference's catalog: the hosted models accept 98,304 input tokens inside a 163,840-token window and cap output at 8,192, so Hermes compresses before an oversized-input refusal; models that cannot call tools are left out of the model list.