Skip to content
navoka
ConnectStart a chat

You set the limits. It stays inside them.

A budget, a threshold, an allowlist and a kill switch, written once. Inside them Navoka pays. Outside them it asks you.

Start a chatHow it's secured
Your rulesActive
Monthly budget
€500
Hard ceiling on everything the agent spends in a month
Approve above
€250
Any single payment above this is held for you
Per-transaction ceiling
€600
Nothing above this, even with approval, without raising it here
Velocity
10 payments / day · €800 / day
Count and value per window
Countries
ES · GB · US · AE
Card and transfer destinations
Kill switch
Stop everything
Freezes cards, pauses rules, unlinks assistants

The rules, on one screen.

This is what "give my agent €500 a month" turns into. Every field is yours; none of them is the agent's.

Your rulesActive
Monthly budget
€500
Hard ceiling on everything the agent spends in a month
Approve above
€250
Any single payment above this is held for you
Payees and merchants
Iberdrola · Amazon · Booking.com · Farfetch · Spotify · Netflix · Anna Reid · Studio Marr
Anyone not on the list is held, whatever the amount
Per-transaction ceiling
€600
Nothing above this, even with approval, without raising it here
Velocity
10 payments / day · €800 / day
Count and value per window
Countries
ES · GB · US · AE
Card and transfer destinations
Categories
Groceries, travel, software, utilities · not gambling, not cash
MCC allow and deny
Mode
Approve each ○ / Run inside the limit ●
Whether in-limit payments still wait for a tap
Card
Agent debit card ••4412 · frozen by kill switch
The card every rule above is checked against, at authorization
Kill switch
Stop everything
Freezes cards, pauses rules, unlinks assistants

Change any of these in the chat with a confirmation, or here. The agent can read them. It cannot write them.It cannot change its own limits

One instruction, a month of payments.

The instruction
Give my agent €500 a month and let it handle the small stuff.
Set: €500 a month, approve anything above €250, allowed merchants and payees as listed. I'll pay inside that without asking and hold everything else. Change it any time — I can't.
Then, over the month
LineOutcomeRule
Iberdrola€84.20Paidallowlisted · under €250
Spotify€10.99Paidsubscriptions card · allowlisted
Amazon · Anker 737€89.99Paidallowlisted · under €250 · agent card
Booking.com · 2 nights€278.00Held → approved by youabove €250
Studio Marr€1,250.00Held → approved by younew payee · above €250 · over the €600 ceiling → ceiling raised by you to €1,300 first
Unknown merchant€39.00Held → declined by younot on the list
14 payments€412.30 of €50011 ran inside the rules3 approved by you1 declinednothing left the account outside your limits

Approve each, or run inside the limit.

Two modes, set per rule, not per account. Start cautious; loosen the rules you trust.

Approve each

  • Every payment waits for a tap, even inside the budget.
  • Default for a new rule.
  • Good for: new payees, travel, anything one-off.

Run inside the limit

  • Payments inside the budget, under the threshold and on the allowlist go through; the rest wait.
  • Switch it on per rule when you've watched it work.
  • Good for: bills, subscriptions, top-ups, payroll batches you've approved before.

Whatever the mode, a new payee, an amount above the threshold, a blocked category or a new country is always held. Mode changes what happens inside the rules, never outside them.Where this applies

The unattended case.

The boring payment is the one that proves the model. Nobody wants to approve an API top-up at 3 a.m.

API credits, unattendedthree weeks
Top up my Anthropic API credits when they drop below €20, up to €200 a month.
Rule set: Anthropic, top up €40 when the balance is under €20, €200 monthly cap, card-on-file, mode "Run inside the limit". No approval inside the cap. I'll post each receipt.
Topped up €40 at 03:12 · that's €160 of €200 this month · one more top-up and I'll hold the next for you.

  • Merchant rule

    Anthropic only

  • Per-transaction ceiling

    €40

  • Monthly cap

    €200

  • Velocity

    max 1 per day

  • Card-on-file

    The agent's own debit card on file at Anthropic, so no 3-D Secure after the first authorization.

Stop everything and show me what it did.

Stop everything and show me what it did.
Stopped. Card ••4412 frozen, rules paused, assistants unlinked. Nothing goes out until you say so. This month: 14 payments, €412.30, all inside your rules — 11 under the line, 3 approved by you. Each one shows the rule, the approver and the reason.

  • Revoke

    Card ••4412 · frozen · 16:04 · Rules · paused · Claude · unlinked · Telegram approvals · still yours.

  • The ledger

    Every row carries four things at the moment of payment: what (Amazon, €89.99), rule (allowlisted, under €250), approver (rule / you at 16:03), reason (your instruction, verbatim: "the Anker power bank, under €120").

  • The dispute pack

    One tap bundles the instruction, the rule, the approval record, the receipt and the timestamps into a single file, ready for a merchant, a card scheme or your accountant.

The reason is captured when the payment is made, not reconstructed afterwards. That's what makes the record usable in a dispute — and what makes your agent's books reconcile without guessing. Unpause: "Unpause" or the button. Rules resume exactly as they were; nothing is re-approved on your behalf.It cannot change its own limits

It cannot change its own limits.

The rules live outside the agent. They are not a tool it can call, a setting it can read-write, or a message it can send itself. Navoka can read the mandate to know what will be held; only you can change it, from the limits screen or in the chat with an explicit confirmation from you. A connected assistant has no path to the mandate at all.

  • The set_policy tool is holder-only;

    an agent or assistant session calling it is refused and the attempt is logged.

  • Approve and Decline buttons are bound to your identity (Telegram account, web session),

    never to the agent's.

  • Raising a ceiling mid-payment (as above) is a separate, logged action by you,

    then a fresh approval.

Where this applies.

Shopping

Merchant allowlist, per-purchase ceiling, daily velocity, all at authorization.

See it on Home

Bills

Allowlist, per-payee ceiling, new-payee hold.

See it on Home

Payments

Per-transfer limit, countries, velocity.

See it on Home

Payroll

Batch approval, per-invoice ceiling, monthly cap.

See it on Home

Trading

Monthly cap, per-order ceiling, asset list.

Regular investing

Ceilings, and how many agents and cards you can run, follow your Wirex One tier. The rules themselves work the same on every tier.

Questions.

How do I stop an AI agent from overspending?
Give it rules it can't change. In Navoka that's a monthly budget, an approve-above threshold, a payee and merchant allowlist, per-transaction and velocity ceilings, a country list and blocked categories. Payments inside the rules go through; anything outside is held for your approval. The agent reads the rules; only you can write them.
What happens when it tries something outside the rules?
It doesn't happen. The payment is prepared and held, and you get a message saying exactly why — "new payee", "above €250", "not on the list", "blocked category". You approve or decline. A decline costs nothing: the authorization never happens and nothing has left the account.
Can the agent change or remove a limit?
No. Limits are set outside the agent, on the limits screen or in the chat with a confirmation from you. The agent can read them so it knows what will be held; it has no tool that writes them. A connected assistant has no access to the mandate at all.
What does "stop everything" actually do?
Immediately: the cards are frozen, every rule is paused, and connected assistants are unlinked. Nothing can leave the account until you unpause. Then you get the month's ledger, each payment with the rule it ran under, who approved it and the reason you gave. Unpausing restores the rules as they were.
What if it pays the wrong thing?
You'll know why it happened: every payment carries your instruction, the rule and the approval. One tap builds a dispute pack — instruction, rule, approval, receipt, timestamps — for the merchant, the card scheme or your accountant. And you tighten the rule so it doesn't happen twice.
Does "approve each" defeat the point?
It's the starting point, not the end state. New rules default to approving every payment so you can watch the agent work. When a rule has behaved — bills, subscriptions, a top-up — switch that rule to "run inside the limit". You end up with a handful of trusted rules running unattended and everything else asking.

Set your limits.

Budget, threshold, allowlist, kill switch. Two minutes, once.

Start a chatConnect your assistant