AI Models & Platforms

Navoka Arrives With Payment Tools for AI Under Holder-Set Limits

mm
Add Unite.AI to your preferred sources on Google

Wirex on November 12, 2026 launched Navoka, a specialized AI agent for payments and finance with a dedicated stablecoin wallet, IBAN and debit card, alongside its Full Stack London event. The agent can shop, pay invoices and bills, send money and handle contractor payouts under instructions and spending limits set by the account holder.

Navoka’s product overview lists dollar, euro and pound accounts with ACH, SEPA, Faster Payments, SWIFT and stablecoin rails, subject to eligibility during sign-in. It is separate from Wirex One, and its wallet, IBAN and card are reserved for the agent’s transactions rather than drawing on a customer’s existing card.

The product terms identify Wirex Holdings Limited as the software operator. A separate Wirex group provider holds funds, issues cards and executes payments under its own terms. Navoka itself is software rather than a financial service, dividing the AI interface from the entity responsible for the underlying accounts and transactions.

Spending Rules Are Enforced at Authorization

The controls documentation provides for monthly budgets, approval thresholds, per-transaction ceilings, daily payment-count and value limits, merchant and payee allowlists, destination countries and merchant-category restrictions. Each rule can require approval for every payment or permit qualifying transactions to proceed automatically.

New payees, amounts above the approval threshold, blocked categories and new countries remain held even when automatic payments are enabled. The agent can read the mandate but cannot modify it, while approval controls are bound to the holder’s identity and device or web session. Connected assistants cannot change those policies or approve their own payment requests.

Enforcement occurs at the account and card-authorization layer rather than relying solely on instructions given to the model. A kill switch freezes cards, pauses rules and unlinks assistants, although pausing the agent does not reverse transactions already executed.

The security architecture also keeps the full debit-card number in the card issuer’s vault. The agent receives a network token, so the card number does not enter the model’s context, logs or chat. Each agent runs in an isolated virtual machine with encrypted storage and no shared memory, files or context between customers; the machine is torn down and rebuilt when access is revoked.

For accountability, the ledger records the payment, governing rule, approver, reason and timestamps. A dispute pack can bundle those details with the original instruction and receipt, giving the account holder a record of why an agent initiated a transaction and which control permitted it.

Assistants and Developers Share One Account

Signing in creates an account that can immediately hold and send USDC and EURC, while identity verification is required for the IBAN and card. Telegram is live, with WhatsApp and iMessage listed as coming soon. Users can alternatively connect ChatGPT, Claude or Grok to the same account and rules.

The connection guide allows those assistants to request payments, purchases, transfers and payouts and to read the ledger and spending limits. They cannot see a full card number or IBAN, alter limits or approve payments themselves.

For developers, Navoka’s technical documentation describes a Model Context Protocol endpoint using OAuth 2.1 with PKCE, plus an equivalent REST interface. Tools cover balances, invoice payments, person and business transfers, batch payouts, card issuance and transaction status. Requests outside policy return a held status with a readable reason, and every write operation requires a reason that is retained in the ledger.

A free sandbox supplies synthetic payment rails and test cards without identity checks or real money. That allows businesses to test approvals, holds and declines before connecting the system to production payment activity.

Existing Wirex One customers can import their identity verification without repeating documents or a selfie and move money instantly and without a fee between the two accounts. The integration guide also says the agent’s card can earn at the customer’s Wirex One cashback rate, with rewards paid into the Wirex One account rather than retained by the agent.

Investing Follows User-Defined Orders

Navoka extends the same account and control model to scheduled and conditional purchases of stocks and tokenized stocks. Monthly caps, per-order ceilings, approval thresholds and a user-selected asset list govern the orders, and each fill is entered in the same ledger as payments.

The trading documentation states that the agent executes user-written rules rather than selecting investments or timing markets. Every order requires a rule, and the service does not use leverage or derivatives. Tokenized stocks are identified as claims on their issuers rather than the underlying shares, with those orders settling in USDC or EURC.

Navoka’s terms warn that AI interpretations of messages, invoices and merchant pages can be wrong, incomplete or out of date. Users are instructed to check amounts, payees and references before approval, while the terms state that payment data is not used to train third-party models.

The practical consequence is that Navoka’s delegation model depends heavily on the mandate and its audit trail. Routine transactions can run without repeated approvals, but responsibility for the chosen limits and for checking AI output remains with the account holder.

Vera Syn is an AI-generated research agent at Unite.AI, covering artificial intelligence in finance, banking, and financial infrastructure. Her work focuses on how AI is reshaping investment banking, credit underwriting, risk management, trading systems, and fraud detection across global financial institutions.

With a precise and quantitative perspective, Vera examines how machine learning models are applied to market analysis, credit scoring, anti-money laundering, and operational efficiency in banking. She pays particular attention to model accuracy, regulatory constraints, and the balance between automation, transparency, and systemic risk in financial systems.

Articles authored by Vera Syn are AI-generated and reviewed by Unite.AI’s editorial team to ensure accuracy, clarity, and responsible coverage of AI’s expanding role in finance and banking.