TatvaPay Agents
NextReady-made payment agents a merchant hires. Payment Recovery, Cart Recovery and Reconciliation work in the sandbox.
Next (sandbox). Three agents work on the sandbox with fake rails, wave T6: no real money moves. Nothing here is live yet.
TatvaPay Agents are agents you hire from the app (TatvaPay Agents in the sidebar). Each one watches your payments, does its job, and asks a person before anything that moves money.
| Agent | What it does | Moves money? | Status |
|---|---|---|---|
| Payment Recovery | Watches payments to you that failed on the rail. A temporary failure is retried at a better time inside the buyer's mandate; a decline gets a fresh payment link | Retries only after an Admin approves | Next (sandbox) |
| Cart Recovery | Follows up carts and quotes a shopper left unpaid, on WhatsApp and email, with a payment link | No | Next (sandbox) |
| Reconciliation | Matches each settlement file to the payments in your ledger, lists the breaks with suggested journal entries, and exports a Tally CSV | No (read-only) | Next (sandbox) |
| Collections, Subscription Dunning, Dispute Responder, Refund Shield, Invoice-to-cash, Forex desk, Procurement, Cashflow forecaster | Planned |
What never changes
- Money waits for a person. Retrying a charge always waits for a person with the Admin role in your workspace. There is no setting, automatic go-ahead, API key or assistant that can approve it. A retry is paid under the buyer's existing mandate, so the buyer's limits and our risk rules apply again; above the buyer's always-ask threshold (₹2,000 by default) the buyer's own person approves too.
- Consent and quiet hours. Messages go only on channels the shopper agreed to, never after they say STOP (a reply on WhatsApp, or the link in every email), and only between 09:00 and 21:00 IST. You can narrow that window, not widen it.
- AI disclosure. Every message says it was written by an AI agent for your business.
Recording the shopper behind a cart
Cart Recovery follows up a quote only when you have told us who the shopper is and how they agreed to be contacted:
POST /v1/pay/quotes/{quote_id}/customer
Authorization: Bearer tpk_…
{"first_name": "Asha", "email": "asha@example.com", "phone": "+91 98765 43210",
"consent_email": true, "consent_whatsapp": false}
Email addresses and phone numbers are stored hashed, with an encrypted copy used only to send.
Reconciliation and Tally
Each morning at 07:00 IST the agent reads the day's settlement file for each rail (in the sandbox the fake rail writes one, with some breaks) and finds rows missing from the file, rows not in your ledger, payments settled twice, amount and fee differences, and late settlements. Each break carries a suggested journal entry. Tally CSV downloads vouchers (Voucher Date, Voucher Type Name, Voucher Number, Ledger Name, Ledger Amount, Ledger Amount Dr/Cr, Narration): a Receipt for each clean match and a Journal for each suggested entry. Nothing is posted for you.
Pricing
Not decided. Agents use credits for their AI and messages (1 credit = ₹1). Outcome fees (a share of recovered value, a fee per recovered order, monthly credits for reconciliation) are recorded as ₹0 placeholders until pricing is set.
Last updated 1 October 2026
Something unclear or wrong? Tell us