Agentic payments with permissions and approvals. Agents ask. Controls decide.

What Flominzo AgentPay is

Flominzo AgentPay is payment permissions and approvals for AI agents. It is an agentic payments product that lets people, businesses, and AI agents initiate payments in natural language, and turns each request into a structured payment instruction that the deterministic payment core executes under mandates, limits, and approvals. It is built for fintechs, wallets, remittance companies, and businesses that want conversational or agent-initiated payments without giving an AI model control of money.

The principle underneath is simple: AI understands and orchestrates intent; deterministic systems control money. Every agent-initiated payment is then reconciled like any other: matched or explained, with evidence. For the controls in detail, see AI agent spending limits and approvals and the free AI agent payment policy template.

AgentPay early-bird pricing

$3,999 setup + $999 a month, for the first 20 customers. The setup covers integration with your providers and bank or settlement files by the Flominzo team, every payment reconciled to closure, and the operations console. Month to month, no minimum term.

From a sentence to a controlled payment

1. Request: A person, a business copilot, or an external AI agent asks for a payment in natural language.
2. Intent understanding: The beneficiary, amount, currency, timing, and purpose are resolved. Anything ambiguous is asked back, not guessed.
3. Structured payment intent: The request becomes the same canonical intent every other Flominzo product uses.
4. Validation: Beneficiary, limits, balance, and permissions are checked.
5. Confirmation or mandate: The person confirms, or a pre-approved mandate covers the payment. Anything outside the mandate is held for a person.
6. Deterministic execution: The payment runs through Send, Payouts, Business, or Bills, then through reconciliation to financial closure.

Every Flominzo product runs through the same lifecycle: intent, plan and execution, rail adapters, events and ledger, reconciliation, and financial closure. Explore the architecture.

What you can ask

Mandates, limits, and approvals

A mandate is the standing authority an agent acts within: who or what it may pay, how much, how often, and when a person must confirm. AgentPay checks the applicable mandate before any instruction reaches execution.

• Limits per payment, per day, and per beneficiary.
• Approval requirements by amount, beneficiary, or payment type.
• Idempotency, so a repeated request does not create a second payment.
• State machines, ledger controls, routing policies, reconciliation controls, and financial closure rules that no agent can bypass.

AI agents for supplier payments and payroll

Supplier payments and payroll are where agent-initiated payments make most sense first. The payees are known, the amounts follow approved invoices or a payroll file, and every payment can be reconciled against bank evidence.

• Invoice runs: a business copilot or an external agent prepares a run from approved invoices, for example “pay all approved invoices due tomorrow, but nothing above £10,000 without asking me”. Flominzo AgentPay turns the request into a structured instruction and checks it against the mandate before anything executes.
• Payroll and contractor payments: the agent assembles the batch; the approvals and limits in the mandate decide what is released, and Flominzo Business executes and tracks every item as its own payment.
• What stays with people: agents never, on their own, move money, change limits or routing policy, or close exceptions. Payments that need approval under the mandate wait for the people you name.
• Proof per item: each payment in the run is reconciled on its own, so finance sees what was paid, what is still processing, and what needs attention.

For the controls in detail, see AI agent spending limits and approvals.

What agents can and cannot do

Agents read, interpret, draft, and explain. They can request a deterministic action through the same controlled commands a person would use. Agents never, on their own:

• move money or change a payment, posting, or ledger entry;
• resolve an unknown outcome by assumption or override idempotency;
• change limits, routing policy, or reconciliation rules;
• close an exception or accept a settlement.

Every tool call an agent makes is recorded as an event with the agent as the actor.

Channels

AgentPay is channel-independent. A messaging app such as WhatsApp, a mobile or web app, voice, an AI assistant, a business copilot, or an external AI agent all resolve to the same structured intent. The channels enabled for you are agreed for your deployment.

100% reconciliation built in

Every payment ends either matched against independent evidence (provider, settlement file, bank statement) or as an open exception with an owner and a reason - none is silently assumed paid.

An agent’s recommendation is not financial proof. Every payment an agent starts goes through the same reconciliation as any other, so you can show exactly what an agent asked for, what was approved, what executed, and what the provider and bank recorded.

Who AgentPay is for

• Fintechs and wallets adding conversational payments.
• Remittance companies offering transfers started in a chat.
• Businesses using copilots to prepare and release invoice runs.
• Platforms that let external AI agents pay, under mandates they control.

Markets

Rails, statement formats, and regulation differ in each market. Flominzo is connected through rail adapters to the providers you use; markets, rails, and activation are agreed for your deployment.

Related products

Questions

Related guides