Questions banks ask.

Tuvel is an agentic banking platform for banks, NBFCs, wealth managers, insurers and fintechs. It gives your staff AI agents on every desk and your customers an AI money advisor, with one control plane over all of them, in your bank’s own private copy.

Being deployed with a Fortune 500 US bank and one of India’s three largest banks.

Three ideas built into the platform. Multiplayer AI: staff and agents work the same case in one room, on one record. Earned autonomy: every agent holds a level and earns wider autonomy from its own record, level by level, with a person deciding each step up and a nightly check that drops a level when the work slips. The Control Plane: one place where every agent’s authority is set, watched and recorded.

Tuvel Advisor, the customer front door. Tuvel Agent Card, a prepaid card only the advisor pays from. Tuvel Agent Teams, 99 staff agents across 14 desks. Tuvel Rooms, where staff and agents work a case together. Tuvel Control Plane, which sets what every agent may do, spend and see. Tuvel Launch, which builds more front doors on the same agents under your brand.

They read documents and your knowledge base, prepare files, draft notes, letters and analyses, track deadlines and propose next steps, each on its own desk: lending, collections, fraud and disputes, KYC and onboarding, compliance, trade and payments, wealth and treasury, service, and legal and bids.

People. Every agent action is tagged: the agent does it, proposes it, or leaves it to a person. Sanctions, approvals, payments, filings and anything a customer sees are a person’s decision, approved with one button on a card.

Level 1 explains, level 2 drafts with a tap on every money action, and level 3 prepares actions ready for one tap. At levels 4 and 5 the advisor pays known bills and merchants from the agent card, within both the customer’s and the bank’s limits, then tells the customer; autopay and monthly investing at level 5 are on the roadmap. Until the bank saves its rules, nothing runs alone.

Credit applications, balance transfers, redemptions, lump-sum investments, loan prepayment, any payment to a person and any payment to a new merchant, at every level. Each tap confirms one exact amount to one exact payee.

Tuvel calls hosted language models through their APIs and has no model of its own trained on customer conversations. Personal identifiers are masked before text reaches a model, and PINs and one-time codes stay out of it. Money maths is done in code.

In a private copy for each bank, with its own database and one fixed outbound IP address the bank can allowlist, inside the bank’s private VPC on Google Cloud, AWS or Azure, in the region the bank chooses. An open-weight model hosted in India keeps every AI model call inside India.

Through an OpenAPI description or an MCP server for each system, granted to one agent at a time and revoked in one click, over mutual TLS, signed requests or OAuth, from one fixed outbound IP address. Connectors for your core banking, loan, CRM, KYC, payment and document systems are built as part of the pilot.

English and ten Indian languages through Sarvam, plus several international languages, by voice or text.

Yes. Describe the agent in plain English in the console: the builder asks what it needs, drafts the agent and grants only the tools listed for it. Your agents see only the details they were granted, pass automated checks from sandbox to staging to live, and only a bank admin puts them live.

A prepaid card only the advisor pays from, loaded by the customer and never drawn from savings, with a default of $120, payments to people capped and always tapped, and a one-tap freeze. It runs in Tuvel today, and is issued on your own prepaid programme in the pilot. See Tuvel Agent Card.

Every turn, every tool call and every gate decision, so your team can see why an action ran, waited or was refused. Calls to bank systems are kept in a tamper-evident, hash-chained trail, with a signed export of those calls your SIEM can ingest and verify.

Our Responsible AI page maps the design to RBI’s published rules line by line, from FREE-AI to the Authentication Directions, for your risk and compliance teams to review.

The same page maps the design to SR 11-7 and OCC 2011-12, the NIST AI RMF, ECOA and Regulation B, Regulation E, UDAAP, the UK Consumer Duty, the PRA’s SS1/23 and the EU AI Act. Each line is designed to support the rule, with the controls behind it: an agent inventory, levels, checks before go-live, the record, a kill switch, maker-checker and your rules first.

We start with your desks, customers and goals. Then we show Tuvel working: the agents on a desk, a room, the control plane and the customer advisor. We finish on fit: which desk or journey first, and what a pilot looks like.

Six to eight weeks on one desk or one journey: scope and success measures, setup in your private copy, a sandbox run, then a small group of staff or customers, and a rollout plan. See the pilot page.

Tuvel is priced for each bank. The price is shaped around the desks, front doors and volumes in scope, agreed after a walkthrough and set out in your pilot plan.