nc://coreweave

CoreWeave Inference: CoreWeave Serverless Inference, its endpoint and its live model catalog, as a Nodecode provider. A Nodecode rewrite of coreweave/serverless-inference-for-hermes by coreweave.

With it, /connect offers CoreWeave Inference, /models lists the models your key can use, and a turn on a CoreWeave model goes to https://api.inference.wandb.ai/v1 over the OpenAI chat lane, as the original sends it:

• the provider's name, endpoint and key variable are the cell's catalog row, so nothing has to be typed in as a custom endpoint
• the model list is CoreWeave's own: /coreweave reads GET /models with your key and puts what it answers in /models. The original's provider ships no fallback list, and neither does this one
• the key is COREWEAVE_API_KEY, the variable the original reads; a key /connect saved answers first, and no other provider's key is ever sent to CoreWeave

See NOTICE.

Get it

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

Set it up

Make an inference-enabled W&B API key at https://wandb.ai/authorize and save it with /connect, or set COREWEAVE_API_KEY in the environment Nodecode runs in. A saved key wins over the variable.

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

{
  "coreweave": {
    // where CoreWeave Serverless Inference is served; the chat lane
    // appends /chat/completions
    "base_url": "https://api.inference.wandb.ai/v1"
  }
}

Use it

/coreweave            # read CoreWeave's model list and put it in /models
/coreweave kimi       # only the models whose id carries "kimi"

Then pick a model in /models and talk to it as usual. An empty or failed list is not proof that a model is unavailable: adding one by hand under providers.coreweave.models works the same way it does for any provider.

MIT licensed.