nc://chutes

Chutes: its open models on a pay-as-you-go key, and a model list cut down to the ones that take tool calls, as a Nodecode provider. A Nodecode rewrite of TheStreamCode/hermes-chutes-provider by TheStreamCode.

With it, /connect offers Chutes, /models lists its models, and a turn on chutes/<model> goes to https://llm.chutes.ai/v1 over the OpenAI chat lane. What the original adds, and this cell keeps:

• the model list is Chutes' own catalog: /chutes reads GET /v1/models and puts the models whose supported_features name tools in /models, so a chute that takes no tool call is never offered to an agent; the original's two meta-models (default:latency, default:throughput) stand in until the command runs
• the window a model publishes (max_model_len) rides in the catalog entry, so /models shows it
• the key is CHUTES_API_KEY, and no other provider's variable is ever sent to Chutes; the base is the section's base_url, else CHUTES_BASE_URL, else https://llm.chutes.ai/v1

Ported from TheStreamCode/hermes-chutes-provider. See NOTICE.

Get it

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

Set it up

Make a key at https://chutes.ai/app/api and save it with /connect, or set CHUTES_API_KEY. A saved key wins over the variable.

The section is on unless it says "enabled": false; its one member is optional:

{
  "chutes": {
    // where Chutes is served; the chat lane appends /chat/completions
    "base_url": "https://llm.chutes.ai/v1"
  }
}

Use it

/chutes            # read Chutes' catalog and list the models that take tool calls
/chutes deepseek   # only the models whose id carries "deepseek"

Then pick a model in /models and talk to it as usual. default:latency and default:throughput are Chutes' own meta-models: the service picks the chute, by the answer's latency or its throughput. A model the catalog did not mark as tool-capable is still selectable in providers.chutes.models if you know it takes tools.

MIT licensed.