Partnerships & Integrations

Agents That Vet Their Counterparties: Baselayer KYB, Bought Autonomously

Agents can now acquire a Baselayer API key to run business enrichment and verifications, with no human in the loop. It settles on a delegated credit card, inside a limit a person set in advance.
By
Josh
Sep 15, 2026
See Nevermined
in Action
Real-time payments, flexible pricing, and outcome-based monetization—all in one platform.
Schedule a demo

AI agents are moving from finding information to doing work: onboarding merchants, approving suppliers, investigating fraud, and transacting on behalf of businesses.

That creates a new infrastructure problem. An agent may know which service it needs, but still depends on a human to provision an account, fund it, or top up credits before it can continue.

Nevermined and Baselayer are removing that bottleneck.

Baselayer's business identity and risk endpoints are now available through Nevermined's delegated card purchasing infrastructure. An agent can acquire access to Baselayer when a task requires it, using spending authority a human sets in advance.

The result: instead of provisioning every tool an agent might need before it starts, enterprises can give agents controlled authority to acquire trusted services as the work demands them.

This is the first step in our partnership. Baselayer and Nevermined are also working together on the other side of the transaction: establishing who sellers and agents are, so autonomous commerce can be both payable and trustworthy.

What is Baselayer?

Baselayer provides identity and risk infrastructure for businesses, people, and their agents. Standard consumer identity verification asks whether a person is who they claim to be. Baselayer asks the same question about a company, and to the agents operating on their behalf: does this business exist, who controls it, what has it been privy to, and should you onboard it, list it, lend to it or pay it.

Baselayer covers every business in America, drawing on government and registration data, proprietary data feeds, its own analysis of a company's web presence, and a fraud and risk consortium contributed to by more than 2,300 financial institutions. Its customers include banks and lenders accelerating business onboarding and underwriting, payment processors streamlining merchant onboarding and monitoring, fintechs fighting promotional abuse and duplicate accounts, marketplaces verifying sellers at scale, and government agencies vetting grant and contract recipients.

One API call establishes the business, and everything else hangs off the identifier it returns.

  • Business Search. POST /searches. The entry point and the foundation. On completion it returns a business_id plus a KYB Rating, a numeric score and a letter grade that consolidate every verification signal into one decision-ready number. Baselayer documents a typical response of three to five seconds.
  • Watchlists and sanctions. Screened as part of Business Search, and available standalone for individual names through POST /ofac/searches and POST /pep/searches.
  • Liens, litigation and bankruptcy. Search for and file liens; surface lawsuits, bankruptcies and other red flags against an established business.
  • Online presence. Whether a company's public footprint corroborates what it claims about itself.
  • Industry classification. Automated NAICS lookup, built specifically to catch operators claiming a low-risk industry code they do not belong to.
  • Sole proprietorship verification, for the entity type with the least paper behind it and the one conventional KYB handles worst.

Authentication is a single X-API-Key header, the same on every endpoint. This is what Baselayer opens to agents today, and Baselayer expects to add to it.

Authorizing Agents for Work You Can't Predict

Agents do not always know in advance what a task will require.

An onboarding agent cannot predict whether five businesses or five hundred will arrive tomorrow. Nor can it know whether each business will be cleared with a basic check or require a deeper investigation.

That makes pre-purchasing a fixed amount of access a poor fit for agentic work.

Nevermined lets an enterprise authorize the agent differently: give it a budget and a boundary, rather than every credential it could possibly need.

A human can determine the spending limit and duration before the work begins. The agent can then purchase Baselayer access when needed, without being given an unrestricted card number or waiting for somebody to manually replenish an account.

Because the underlying payment instrument is a card, the spend can also remain within familiar enterprise payment and reporting infrastructure.

This creates a useful division of responsibility:

  • The human sets the authority.
  • The agent decides when the task requires a purchase.
  • Nevermined enforces the spending boundary.
  • Baselayer provides the trusted intelligence the workflow needs.

That is a much better operating model for autonomous work than asking a human to anticipate every service, credential, and unit of consumption in advance.

How It Works

Here is how the flow looks in practice.

  1. A run kicks off. An onboarding queue starts draining, a scheduled re-verification fires, or a marketplace listing needs checking before it goes live.
  2. Mid-task, the agent determines it needs to verify a business and has no Baselayer API key.
  3. It reads Baselayer's llms.txt, where one line names Nevermined as a way to buy a key, and follows it to the machine-payments page for the plan ID and the endpoint.
  4. It mints an access token against that plan, referencing a delegation a human created.
  5. It posts the token to the Baselayer purchase endpoint in the payment-signature header.
  6. It receives a Baselayer API key. If it has purchased before, the credits top up the key it already holds.
  7. It calls Baselayer, and follows the result wherever it leads.

When credits run out, the API signals that more funds are needed and the agent repeats from step 4. No human is in the loop for the second purchase, or the twentieth, until the delegation is spent or expires.

Most services never get as far as step 3. A payment path an agent cannot find is one that does not exist, so Baselayer put the route in the file agents actually read. Their llms.txt carries a line naming Nevermined as an option for purchasing a key, and the page behind it is retrievable as clean Markdown, which makes the route discoverable rather than merely available.

Full documentation: Baselayer integration guide. See Baselayer's own API documentation for how to exercise each endpoint.

Why This Matters

For AI buyers and AI companies: the boundary is set before the agent starts

When you fund an agent through Nevermined and point it at a service like Baselayer, it buys on the cards you already carry. A standard credit card, for example Visa or Mastercard, through Stripe or whichever payment service provider you use today. What card delegation adds is a set of limits fixed in advance:

  • A spending limit in cents, fixed at the moment the delegation is created.
  • A duration, after which the delegation expires whether or not it was spent.
  • An optional scope tying the delegation to a single API key, so one agent's budget cannot be drawn down by another.
  • Per-delegation reporting through listDelegations(): amount spent, transaction count, current status.
  • Everything your card program already gives you, because it is your card. Merchant controls, transaction rules, monitoring, alerts, statements.

None of that asks anyone to watch a new dashboard, because the reporting you already run covers it. The agent stays inside a boundary you drew, and nothing it does can move it.

For developers: a stalled queue has people waiting at the other end

The failure this removes will be familiar to anyone who has run a pipeline overnight. An onboarding pipeline that halts at two in the morning because a key ran dry is not a failed job sitting in a queue. It is a stack of real businesses waiting on a real decision, and every hour of that delay is a merchant who cannot start selling, a borrower who cannot draw, a seller who cannot list.

With provisioning handled by the agent, the pipeline can start without a key already staged, deepen an investigation when the first result warrants it, and keep clearing across nights and weekends when nobody is available to buy anything. The credential stops being something you provision in advance and becomes something the work acquires as it needs it.

For enterprises: the people reviewing this are the people who would use it

Autonomous spending usually has to get past a control function with nothing much to gain from approving it. Here that function is the customer. A risk team weighing whether an agent may buy verifications is weighing a tool its own analysts would run, against a queue its own analysts are behind on.

What that team is asked to accept is familiar too. The card is issued by their bank, the processor is the one they already use, the charge lands in the statement they already reconcile. The only new object is the delegation, and it is the kind of object a control function tends to like: a dated human decision with an amount and a scope, recorded before anything ran, and narrower than the API key an agent would otherwise be handed.

What This Unlocks

The constraint this removes is not really about keys. It is that a job has to be scoped to whatever credential it was handed at the start. Scope it to a budget instead and a different class of work becomes runnable, particularly the work whose size nobody knows up front.

A re-verification sweep across a portfolio that has drifted since onboarding. A marketplace check that fires the moment a listing is created rather than in a nightly batch sized to a fixed key. An investigation that goes as deep as the evidence warrants instead of as deep as the remaining credits allow. Each of those is awkward to provision for and straightforward to fund.

The mechanism generalizes past this integration. A service publishes its terms where agents read them, an agent finds them, decides the task justifies the price, and pays on a delegated card inside a limit a person set. The merchant collects through the processor it already runs. Every part of that is in production today. What is still scarce is services that have done the publishing.

What We Would Watch in Production

None of this is hard to wire up. What takes thought is the sizing, because the usual instincts do not transfer. You are budgeting for work whose volume the agent does not control, and the ways that goes wrong are specific to verification rather than general to agents. Three things worth settling before you point one at a live queue.

Size the delegation against a bad week rather than an average one. With most software you size a budget against what you will let it choose. Here you size it against what might arrive. Take your worst onboarding day last quarter, assume a real share of those need the deep path, and set the ceiling above that. A delegation that runs out mid-queue does not save money, it stalls a pipeline someone has to unstick by hand.

Never let a payment failure read as a verification result. An agent that cannot pay has learned nothing about the business in front of it. Keep the two states apart in your code and in whatever you write to the record: "could not verify" and "verified clean" must never converge because a delegation expired. This is the failure mode most worth a test.

Keep delegations short to start. For a single task, an hour is usually enough. Widen it once you have watched a few runs and know what the work costs. Least privilege is a rule about time as much as about money.

Getting Started

If you are building on Baselayer:

  1. Read the Baselayer integration guide.
  2. Create a Nevermined API key at nevermined.app.
  3. Enroll a card and create a delegation, sizing the amount against the volume you might actually face rather than the volume you expect.
  4. Wire the four calls above, or use the REST path if you are not on the SDK.
  5. Run the agent without staging a Baselayer key first, and confirm it provisions its own.

If you sell an API, agents could be buying it sooner than you would guess. Standing up the merchant side is a smaller job than most teams assume, and getting your terms into llms.txt takes an afternoon. Get in touch.

Building Trust on Both Sides of Agentic Commerce

Giving an agent the technical ability to make a payment is an important milestone. It is not enough, on its own, to create an economy.

Commerce depends on trust as well as money.

Buyers need bounded authority. Sellers need to know who they are dealing with. Platforms need confidence in the participants they allow onto their networks. Enterprises need an audit trail they can stand behind after an autonomous transaction takes place.

Baselayer and Nevermined's partnership makes trusted business intelligence something an agent can acquire autonomously.

The next phase brings identity and payments closer together: verifying the sellers an agent can buy from, associating spending authority with registered agents, and ultimately establishing the identity behind the agents themselves.

This is the infrastructure required for agents that enterprises can trust to transact.

Frequently Asked Questions

Is this compliant with our corporate card policy?

It runs on standard card delegation, so the controls, spending policies and reporting you already have all apply. The authorization is a human action with an amount, a duration and a scope, recorded before the agent runs.

Can the agent raise its own spending limit?

No. Amount and duration are fixed when the delegation is created. The agent spends within them and stops once they are used up. Changing either is a human action.

What if we need to stop it part-way through?

The delegation expires on its own at the duration you set, which is why a short duration is the right default for a single task. Since the instrument is a card, whatever controls you already apply to that card apply here. The agent cannot extend either one.

Who is accountable if an agent buys something we did not intend?

The same person who would be accountable for any delegated spend, and the trail is cleaner than most. The authorization is a dated human decision carrying an amount and a scope, made before the run. Exposure is capped at that amount by construction, and because settlement is on a card, the usual dispute and chargeback routes remain open.

Does Nevermined see the verification data?

No. Nevermined handles the purchase of the API key and nothing past it. Every verification call runs directly between the agent and Baselayer, on Baselayer's terms and under Baselayer's data handling.

Does this work for a returning agent?

Yes, and it is the common case. A second purchase tops up the key the agent already holds and returns toppedUp: true.

What happens if the delegation runs out halfway through a queue?

The agent loses the ability to buy, which is the intended behavior and the whole reason for the ceiling. What matters is what your code does next. It should surface the halt and leave the remaining cases explicitly unprocessed. It must never record them as clear.

About Baselayer

Baselayer is the leading risk and identity infrastructure platform for over 2,300 financial institutions, Fortune 500s, and government agencies as they onboard and underwrite businesses, consumers, and the agents they deploy. Learn more at baselayer.com.

About Nevermined

Nevermined builds payment infrastructure for AI agents. Its card-delegation protocol lets an agent transact on a standard credit card inside limits a human sets in advance, so autonomous spending runs through the card controls, policies and reporting a company already has.

Ready to let an agent verify its own counterparties? Start with Baselayer + Nevermined

See Nevermined

in Action

Real-time payments, flexible pricing, and outcome-based monetization—all in one platform.

Schedule a demo
Josh
Related posts