nc://zerosignal

ZeroSignal: private, pay-per-request inference through the local zs-proxy, with a funded wallet as the credential. A Nodecode rewrite of TxnLab/hermes-zerosignal-provider by TxnLab.

ZeroSignal is a pay-per-request, privacy-preserving inference network reached through zs-proxy, a proxy on the same machine that speaks the OpenAI chat wire at http://127.0.0.1:9376/v1. With this cell, /connect offers ZeroSignal, /models lists the models the proxy currently serves, and a turn goes to the proxy:

• the request's reasoning_effort is clamped down onto the levels the serving node declares in the proxy's own /v1/models (reasoning.supported, reasoning.allowed_efforts) — never upward, so a node that declares low/high/max is never sent medium
• a model that declares reasoning.supported: false is sent no reasoning parameter at all, and a model the proxy has not named passes through as it is
• no key is needed: any non-empty value works, because the proxy ignores the Authorization header and the wallet pays

Get it

Tell the agent "add the zerosignal cell", or run nodecode add nc://zerosignal.

Set it up

Start and fund the proxy first on the same machine:

zs-proxy proxy start
zs-proxy fund

See https://docs.zerosignal.ai/using-the-proxy/guides/hermes for installing zs-proxy.

Then pick ZeroSignal in /models. The picker asks for a key; any non-empty value does (zerosignal-local is the one the cell sends when none is saved), and ZEROSIGNAL_API_KEY in the environment works the same. The proxy's address is optional, for a proxy on another port:

{
  "zerosignal": {
    "base_url": "http://127.0.0.1:9376/v1"
  }
}

Use it

Ask for a model — /models lists what the proxy serves, and the curved list in this folder's fallback is what it shows before the proxy answers — then talk as usual. /think offers the rungs the chosen model declares, and a pick outside them is clamped before it reaches the network.

Test

From this repository's root:

.github/test-cell.sh providers/nodecode-zerosignal

MIT licensed. See NOTICE.