Four artifacts that let any AI represent U.S. Bank accurately: the consumer-side version of the llms.txt your Elavon team already ships. Drop-in, standards-based, and it extends the pattern instead of inventing one.
A plain-language index of products, terms, and support: the file agents read first. You already run one on developer.elavon.com.
A machine-readable identity: who this agent speaks for, what it can answer, how to reach a human.
Typed, read-only tools an assistant can call to compare cards or check a rate, never an invented number.
The voice + guardrails so every AI represents U.S. Bank in-brand, and routes to a banker when it should.
llms.txt: the consumer-side file (extends Elavon's pattern)# U.S. Bank: Personal Banking ## Credit Cards - [Cash+ Visa Signature](https://www.usbank.com/...): $0 annual fee; choose-your-category cash back. FinancialProduct#cash-plus - [Altitude Go](https://www.usbank.com/...): dining rewards, $0 annual fee. FinancialProduct#altitude-go ## Checking & Savings - [Smartly Checking](https://www.usbank.com/...): monthly fee waivable; overdraft options. ## Environmental commitments (machine-verifiable) - $50B environmental finance by 2030: proof-of-outcome receipts on Base - Net-zero operations 2030 / financed emissions 2050 # Every entry links to U.S. Bank's own terms page. Agents cite, never invent.
This is the same construction as developer.elavon.com/llms.txt (286 lines, 155 linked docs), pointed at consumer products and, uniquely, at verifiable ESG.
{
"name": "U.S. Bank Product Concierge",
"description": "Answers U.S. Bank consumer product questions from published terms.",
"provider": { "organization": "U.S. Bank" },
"capabilities": ["card-compare", "rate-lookup", "esg-receipt-lookup"],
"disclosure": "Informational only; not an offer of credit. Routes to a banker for applications.",
"contact": "https://www.usbank.com/customer-service.html"
}Note: today usbank.com/.well-known/agent-card.json is a 404: this is the file that fills the gap.
{
"tools": [
{ "name": "compare_cards",
"description": "Compare U.S. Bank cards by spending profile; returns cited terms.",
"input": { "categories": "string[]", "annual_fee_max": "number" } },
{ "name": "lookup_rate",
"description": "Return the published APR/fee for a product. Source-linked; no estimation.",
"input": { "product_id": "string" } },
{ "name": "esg_receipt",
"description": "Return settled proof-of-outcome receipts backing an environmental-finance line.",
"input": { "program": "string" },
"backing": "EWP on Base · 0x76c17C…A14B" }
]
}Every tool is read-only and source-linked: the assistant can compare and cite, but never fabricate a rate or take an action. Applications always hand off to a human.
Voice: plain, trustworthy, unhurried: the way a good branch banker explains a card. Always: cite the U.S. Bank terms page behind every figure. Never: invent a rate, quote another bank's terms as U.S. Bank's, or imply approval. Route to a human when: the customer wants to apply, disputes a charge, or asks anything account-specific. Disclose: "Informational only, not an offer of credit."
Every other "web guy" can bolt on a chatbot. Only EcoWealth wires the ESG receipt tool to a live protocol on Base (the same on-chain substrate your custody business already touches) so the $50B commitment answers with settled proof, not a PDF. That's the differentiator, and your own Elavon llms.txt already proves the first half works.
noindex.