数牛-nex-pro: agriculture vertical preset

A pre-tuned bundle of nex-pro + an agriculture-expert system prompt for Chinese-language farming, supply-chain, and rural-economy questions. Holders of the DNIU holder token get automatic discounts of up to 20% on every request that names this preset.

1. Why a preset?

A preset is a curated (model, system prompt, tool chain, RAG corpus, discount rule) bundle that ships with a single slug. Customers pass the slug as the model field in /v1/chat/completions — everything else is server-side:

• The model name on the wire stays the same as a regular call, so OpenAI SDK clients work without modification.
• The preset's system prompt is injected automatically — the customer's own system message (if any) is preserved and prepended inside, so calling intent stays intact.
• If a RAG corpus is bound, the relevant chunks are retrieved and inserted as additional context before your user message.
• If you hold the required holder token at the required threshold, the bill is automatically discounted.

2. Call the preset

Use the OpenAI Python SDK pointed at NexToken — no other changes:

from openai import OpenAI

client = OpenAI(
    api_key="sk-nex-...",                       # your NexToken API key
    base_url="https://api.nextoken.biz/v1",
)

resp = client.chat.completions.create(
    model="数牛-nex-pro",                        # ← preset slug
    messages=[
        {"role": "user", "content": "海南三月种水稻合适吗？要注意什么？"},
    ],
)
print(resp.choices[0].message.content)

The reply will be grounded by the agriculture-expert framing and will reference public 农业农村部 data where applicable. The customer-facing model name in resp.model is the resolved upstream model (currently nex-pro) — the preset id is recorded in your usage record + audit log for auditing, but isn't echoed in the OpenAI response.

3. Auto-applied discount

The 数牛-nex-pro preset's discount config:

The discount engine runs at billing time, reads your user_holder_balances row, walks the tiered config, and applies the highest qualifying rate. No flag to set, no field to pass in the request — it just works if you qualify.

4. Check your eligibility

The /v1/holder endpoints surface what you have and what you qualify for:

import requests

BASE = "https://api.nextoken.biz/v1"
HEADERS = {"Authorization": f"Bearer sk-nex-..."}

# Tokens you hold
balances = requests.get(f"{BASE}/holder/balances", headers=HEADERS).json()
print(balances["balances"])
# [{"holder_token_code": "DNIU", "balance": "500", "granted_via": "admin", ...}]

# Presets you qualify for + the pct you'd receive
eligible = requests.get(f"{BASE}/holder/eligible-presets", headers=HEADERS).json()
print(eligible["presets"])
# [{"preset": {"id": "数牛-nex-pro", ...},
#   "required_holder_token": "DNIU",
#   "your_balance": "500",
#   "effective_discount_pct": "0.10"}]

5. Audit the savings

Every time the discount engine takes effect, a PRESET_DISCOUNT_APPLIED audit row is written. Read your savings history through:

tx = requests.get(f"{BASE}/holder/transactions?limit=20", headers=HEADERS).json()
for row in tx["data"]:
    saved = float(row["base_amount_usd"]) - float(row["discounted_amount_usd"])
    print(f"{row['applied_at']}  preset={row['preset_id']}  "
          f"−{row['discount_pct']}  saved=${saved:.6f}")

Or open the Holder Portal in the dashboard — the Savings History tab paginates the same data with running totals.

6. Browse other presets

数牛-nex-pro is one of an expanding set of vertical presets. Browse the Preset Store for the full catalogue, or list programmatically:

r = requests.get(f"{BASE}/presets", headers=HEADERS).json()
for p in r["presets"]:
    print(p["id"], "—", p["display_name"], "—", p["vertical"])
# 数椰-nex-pro — 数椰 / Tropical Crop (nex-pro) — 热带农业
# 数牛-nex-pro — 数牛 / Agriculture (nex-pro) — 农业

7. Putting it together

Full minimal example, suitable for an agriculture support bot:

from openai import OpenAI
import requests

API_KEY = "sk-nex-..."
BASE = "https://api.nextoken.biz/v1"

client = OpenAI(api_key=API_KEY, base_url=BASE)

def ask_agri_expert(question: str) -> dict:
    """Ask the 数牛-nex-pro vertical preset, return answer + cost."""
    resp = client.chat.completions.create(
        model="数牛-nex-pro",
        messages=[{"role": "user", "content": question}],
    )
    return {
        "answer": resp.choices[0].message.content,
        "usage": resp.usage.model_dump() if resp.usage else None,
    }

result = ask_agri_expert("海南春季芒果裂果的常见原因？")
print(result["answer"])
print("Tokens used:", result["usage"])

# Look up the actual price paid after discount (asynchronously
# arrives in the audit log a few seconds after the call):
tx = requests.get(
    f"{BASE}/holder/transactions?limit=1",
    headers={"Authorization": f"Bearer {API_KEY}"},
).json()
if tx["data"]:
    last = tx["data"][0]
    print(f"Discount applied: {last['discount_pct']}  paid ${last['discounted_amount_usd']}")

FAQ

Will my own system prompt still be used?

Yes — if you pass a system message in messages, your content is prepended inside the preset's system message rather than being replaced. Order: preset framing first (broad context), your specifics after (specific intent). One system role total.

What if I pass my own tools?

Explicit caller intent always wins — if you pass tools (non-empty), the preset's tool chain is skipped entirely. Pass tools=[] or omit to use the preset's tools (if any).

How is the discount calculated for combined rules?

The discount engine is capped at 0.95 defensively — even a config typo that says "1.5 fixed_pct" can't make calls free. The "Reference" docstring on app/services/discount.py documents the cap and gating logic.

Can I see the system prompt before I call?

Yes — the public catalog endpoint returns it. GET /v1/presets/数牛-nex-pro surfaces system_prompt, tool_chain, and discount_config. Presets are intentionally transparent — there's no proprietary moat in the prompt itself.