nc://aixy

Aixy: an inference gateway as a Nodecode provider, with its live model catalog. A Nodecode rewrite of aixy-tech/aixy-hermes-plugin by aixy-tech.

Aixy serves models over the OpenAI chat wire at one base, with a project key. With this cell, /connect offers Aixy for that key, /models lists the models the gateway itself advertises (asked live, with the key, never a list baked into this folder), a turn on aixy/<model> goes to https://api.aixy-gateway.com/v1/chat/completions, and /aixy answers what the gateway serves and where its key comes from.

Get it

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

Set it up

Make a project key in Aixy, then either save it with /connect (pick Aixy) or set AIXY_API_KEY. The key is a secret in either place and this folder stores nothing.

Nothing else is needed: the section defaults to the public gateway, and base_url is where a gateway or relay of your own is served:

{
  "aixy": {
    "base_url": "https://api.aixy-gateway.com/v1"
  }
}

The base must be HTTPS, or HTTP on this machine's loopback; a base carrying credentials, a query or a fragment is refused at start, because the project key is what the cell sends to it. See config.example.jsonc.

Use it

/aixy          what the gateway serves: each model id and its context window
/aixy key      where the project key comes from, and where to make one
/models        the picker, where one of those ids becomes this session's model

A conversation's requests carry its scope in the X-Aixy-Session-ID header, which is how Aixy Activity groups them. "session_header" names that header when your gateway spells it differently; "none" sends none.

Two things the rewrite leaves to Nodecode rather than carrying over: the plugin's interactive wizard (its work is /connect, /models and /setup here) and its optional MCP OAuth step (Nodecode's own mcp cell adds an MCP server). See NOTICE.

Test

From this repository's root:

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

MIT licensed, as the original is.