Product · Provider Entry

Provider Entry

List Taifoon as an external provider on your inference cloud. One OpenAI-compatible entry, three n8n workflows in the admin shape, the first routes at the catalogue price, and two ways to settle.

An inference cloud routes a model name to a backend. The Provider Entry is one such backend: a base URL, a key in an environment variable, a model list at /models and chat at /chat/completions. Behind it sit every venue the brokerage buys from and the rigs of the provider program, priced from the market, settled through one vault. The cloud approves the routes it wants; each answer carries its venue and its sale id.

Status: quoted. The door answers with a price (the catalogue below, and a 402 with terms on a call without a key). It reads live once an outside buyer's payment is received into the vault. The worked example is NEAR AI Cloud, in its admin shape; the same entry fits any cloud or router with an external OpenAI-compatible backend.

1. The entry

Seven fields. The key behind TAIFOON_API_KEY is minted by the cloud at POST https://coord.taifoon.dev/v1/register and funded by the cloud; its value is never in a workflow or on this page.

In the admin shape

On NEAR AI Cloud the provider registry takes the entry as one object, then approves {upstream_model, model_name} pairs from what /models returns. The model batch of today's admin API takes one row per model; the price fields are the catalogue's number per 1M tokens, scaled to a per-token amount.

{
  "name": "taifoon",
  "config": { "backend": "openai_compatible", "base_url": "https://coord.taifoon.dev/v1" },
  "credential_env": "TAIFOON_API_KEY",
  "models_path": "models"
}

{
  "modelName": "<the id from https://coord.taifoon.dev/v1/models>",
  "providerType": "external",
  "providerConfig": { "backend": "openai_compatible", "base_url": "https://coord.taifoon.dev/v1" },
  "attestationSupported": false, "verifiable": false, "isActive": true,
  "inputCostPerToken":  { "amount": "<usd_per_1m.in  × 1000>", "scale": 9, "currency": "USD" },
  "outputCostPerToken": { "amount": "<usd_per_1m.out × 1000>", "scale": 9, "currency": "USD" }
}

Discovery

curl -s https://coord.taifoon.dev/v1/models | jq '.data[] | {id, owned_by, price: .taifoon.usd_per_1m, available: .taifoon.available}'

At the last read (2026-10-26 14:46 UTC) the catalogue carried 28 models from 8 venues: chutes, deepinfra, groq, ionet, near-ai, phala, taifoon-gpu, venice. A model moves with the market; the price a request pays is the one the catalogue shows at that moment.

2. Who it is for

A cloud whose policy requires a TEE lists only attested routes. This entry is not one. The pool's rigs can join the cloud's own TEE stack later; that is step 6 of the rollout, after paid volume on this door.

3. The three workflows

Import them into n8n (Workflows → Import from URL, or Import from File). Each holds a credential placeholder only. The calls to a cloud's admin API are in the files and disabled; the cloud enables them on its own n8n with its own credential.

4. The first routes

The routes a cloud approves first, each with the catalogue entry that serves it now and that entry's price. Every number is read from GET /models when this page renders; a route with no listed supply prints — and enters the day a venue lists it.

5 routes · read 2026-10-26 14:46 UTC · the full list and every other model: https://coord.taifoon.dev/v1/models

5. The settlement forms

The cloud picks one. Both end in the same ledger: one sale per answer, its legs netting to zero per token and chain, the public summary at api.taifoon.dev/coordination/sale-run.json.

curl -s -X POST https://coord.taifoon.dev/v1/register -H 'content-type: application/json' -d '{}'
curl -s 'https://coord.taifoon.dev/v1/access/quote?usdc=10&chainId=8453'

curl -s -X POST https://coord.taifoon.dev/v1/chat/completions -H 'content-type: application/json' \
  -d '{"model":"deepseek/deepseek-v3.2","messages":[{"role":"user","content":"ping"}],"max_tokens":4}'

The settlement line a cloud posts per attempt is computed in workflow 21: the catalogue price is USD per 1M tokens, so micro-USD per token is the same number; cost = prompt tokens × in + completion tokens × out.

6. The rollout