nodecode-gondola

Gondola: a USDC-settled marketplace for Venice AI inference, as a Nodecode provider whose model picker carries only the chat routes. A Nodecode rewrite of Gondola-Market/hermes-gondola-provider by Gondola-Market.

Get it

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

Set it up

Mint a key at https://gondola-ai.com/wallet?tab=keys (a Gondola key starts with gnd_) and top the wallet up with USDC on Base once at https://gondola-ai.com/wallet. Save the key with /connect (pick Gondola), or set GONDOLA_API_KEY in the environment; a saved key answers first. Each request is metered against the prepaid balance.

The section is optional — the base defaults to Gondola's own:

{
  "gondola": {
    // where the OpenAI-compatible API is served; the chat lane appends
    // /chat/completions, and /models is read here
    "base_url": "https://api.gondola-ai.com/v1"
  }
}

Use it

/models lists Gondola's chat models, and a turn on one goes to <base_url>/chat/completions.

Gondola serves one OpenAI-compatible endpoint in front of Venice's models and routes each request to the cheapest supplier. Its GET /v1/models is a mixed catalog — text, image and video models together, the majority not chat — so the cell keeps only what a turn can run on: an entry whose endpoint is /v1/chat/completions, else one whose modality is text, else one that names neither, which is kept the way the stock picker would keep it, so a catalog shape the cell does not know degrades to "list everything" rather than to an empty picker.

When the catalog is unreachable, the picker falls back to the five tool-calling models the original plugin curates — deepseek-v4-flash, claude-sonnet-5, kimi-k2-7-code, qwen3-coder-480b-a35b-instruct-turbo, venice-uncensored-1-2 — so an offline start still fills it.

Test

From this repository's root:

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

MIT licensed — see LICENSE. Ported from the original plugin; see NOTICE.