Mandates and AutoPay
NextHow an agent pays from a person's bank account inside caps set once, and how TatvaPay mandates sit on top of UPI AutoPay, e-mandates and Reserve Pay.
Coming next. TatvaPay mandates, caps, approvals, the consent centre and signed receipts work in the sandbox (fake rails, no money moves). Bank mandates (UPI AutoPay and e-mandates) are being built with the first planned payment integration, Razorpay, in test mode. UPI Reserve Pay is planned. Nothing is live in production.
A TatvaPay mandate works like the AutoPay a person sets up for a subscription, but for one AI agent. The person authorises it once; after that the agent can pay only inside its limits, and the money stays in the person's bank account until a payment is made.
Two layers
| Layer | What it knows | Who runs it | Status |
|---|---|---|---|
| TatvaPay mandate | Which agent, caps per payment and per period, merchants or categories, expiry, the approval threshold | TatvaPay (rules in code, versioned) | Next (sandbox) |
| Bank mandate | A maximum amount and frequency the bank will debit | NPCI and the person's bank, through a licensed payment aggregator | Next (being built), Planned for Reserve Pay |
A payment goes through only if it fits both. The bank mandate alone cannot tell which agent is paying or for what; the TatvaPay mandate adds that.
The lifecycle
- Draft. An agent (or the app) proposes a mandate: caps, categories, expiry. Agents can ask for one with the MCP tool
request_mandate; they cannot authorise it. - Authorise. The person reads the consent wording and authorises once. The exact wording version and its hash are recorded with the mandate. For a bank mandate, the person also approves it in their UPI app or netbanking, as with any AutoPay.
- Active. The agent pays inside the caps. Every payment is checked by the rules engine and lands in a hash-chained, signed receipt.
- Ends as exhausted (caps used up), expired or revoked (by the person, in the consent centre or, for UPI AutoPay, in their UPI app).
The ₹2,000 rule
Above a threshold, ₹2,000 by default, the person is always asked, whatever the mandate allows. An agent cannot raise the threshold. Merchant agents' retries of a failed charge (Payment Recovery) always wait for a named person on the merchant's staff, at any amount.
The bank rails
| Rail | In one line | Status |
|---|---|---|
| UPI AutoPay | A recurring UPI mandate approved once with the UPI PIN; later debits stay within the approved maximum, usually with a notice before each debit | Next (being built, test mode) |
| e-mandate | The same idea through netbanking or a debit card | Next (being built, test mode) |
| UPI Reserve Pay | An amount blocked in the person's account for an agent; payments are taken from the block as needed | Planned (when the payment aggregator enables it) |
| Card agent tokens | Network tokens scoped to an agent | Planned |
Maximum amounts, pre-debit notices and when extra authentication is needed are as set by NPCI and your bank. They change over time; we do not repeat numbers here.
Mandates from other protocols
Mandates created over ACP (delegated payment on a UPI mandate) and AP2 (open and closed mandates) are mapped onto the same model, carry their source, and show up in the consent centre with one-tap revoke.
Who holds what
TatvaPay holds no RBI licence and never holds funds. Bank mandates and debits are run by licensed payment aggregators (first planned integration: Razorpay). See Trust and security.
Related: Mandates (the data model), Consent centre, How mandates work (plain-language guide).
Last updated 2 October 2026
Something unclear or wrong? Tell us