Sahaya is ambient AI for money. Connected once to your accounts, email and calendar, it works in the background, speaks your language, and acts across India's financial rails with your consent and on your side.
Every family that does well with money has someone they ring before a big decision: the uncle who understands insurance, or the cousin who works at a bank. Most Indians don't have that person. Sahaya is meant to become that person, living in the phone already in your pocket. It speaks your language, knows your numbers, and gets sharper with every conversation.
It isn't another app to open. It works in the background and comes to you first, before you sign, renew or invest.
Fin 1.0 was built on selling products. Fin 2.0 was built on doing transactions. Fin 3.0 is being built on intents. The hard part of managing money was never the transaction. It lives in the intent behind it.
Banks pushed accounts, policies and loans across a counter. The relationship was the product you were sold.
UPI made every payment instant and free. Apps optimised for taps and the transaction feed, but a payment is only the last step of a decision.
"I want an AC before summer." "Can I afford my daughter's college?" An agent that takes the intent, works out every step behind it, and only then touches a rail. This is where Sahaya is built.
The whole complexity of managing money lives in the intent, not the transaction.
Agents now work in the background. They read your email and calendar, watch for changes, and act with your approval. Very few understand Indian financial rails, or speak Tamil or Telugu.
Every conversation teaches the system something. Memory compounds into a moat that's structural rather than a feature.
UPI, Account Aggregator, BBPS, DigiLocker and Aadhaar make up programmable rails that already serve hundreds of millions of people. UPI alone clears 24 billion payments a month, and Account Aggregator has fulfilled 49 crore consent requests.
Big Tech will own general-purpose AI. Regulated verticals need trust, compliance and local context, which global players won't build. That's the white space, and money is the domain where going deep matters most.
India built the most advanced digital financial infrastructure on the planet. Payments clear in seconds, data moves on consent, every biller is a tap away. And yet hundreds of millions of people with bank accounts still can't navigate it on their own.
Sources: NPCI (Sep 2026) · NPCI Bharat Connect · NCFE financial literacy survey · Sarvam AI
The infrastructure is there.
The interface is missing.
Sahaya is the interface.
A Conductor breaks each request into tasks and routes them to the right specialist agents. A Compliance agent runs outside the process and can veto any of them. The user only ever talks to one companion.
Sahaya doesn't ship a fixed app of screens. When you ask for something, it builds a page for that task from a library of safe, typed blocks (amount, plan, comparison, chart, timeline, consent). The page disappears when the job is done, or you pin it.
You see your full money picture before Sahaya asks for anything. No forms, and no upfront KYC.
Every read and every action goes through a signed, purpose-bound consent artifact. There's no bypass path.
Native financial vocabulary, not translated English. "Chit fund", "kharcha paani" and "EMI baaki" are understood in context.
Sahaya is built for the hundreds of millions of Indians who have a bank account and a smartphone, but not the financial literacy, the language access or the time to use the system well.
Every message runs inside an agent harness, every reply is composed from typed, schema-checked blocks, and every financial action passes through consent and compliance on its way down to India Stack, with no exceptions and no back door. Select a layer to inspect it, or trace one of Meena's voice notes through the whole stack.
Three things make Sahaya hard to copy: privacy, consent and memory. Model routing uses three tiers that trade off capability, cost and compliance. The privacy router strips identifiers in the infrastructure layer, before a query can reach a larger model. Type anything below and watch what each tier is allowed to see.
Design rule: Aadhaar numbers, PAN numbers, bank account numbers and phone numbers never leave Tier 1. Infrastructure enforces this, not application code, so no agent prompt can override it.
Earn trust through conversation, then earn permission through demonstrated value — never the other way around. Each consent is purpose-bound, cryptographically signed, time-limited and revocable in one phrase. Try revoking one.
Exactly what Sahaya needs to see.
The single purpose it will be used for.
How to end it, at any time, in your words.
Traditional KYC tells you who someone is legally, and nothing about who they are financially. Sahaya treats every message as a signal. Within a handful of conversations, Sahaya knows more about someone's financial life than a KYC form ever could.
Sahaya becomes the place an Indian household keeps its financial keys: a zero-knowledge password manager and a tokenised card store, alongside reward points, warranties, policies and goals. Because it holds the keys, it can do more than advise. It can find the best price on a phone, an AC or a ticket home, stack every offer, then sign in and pay once you say yes.
Whoever holds the keys makes the purchase.
The core experience is free. Pro (₹99 a month or ₹999 a year) adds tax optimisation, retirement planning, a family dashboard and priority support, and brings Sahaya into the AI tools people already use: Sign in with ChatGPT, an MCP server and APIs. The rest of the revenue comes from being a transparent distribution layer: every recommendation discloses whether Sahaya earns a commission, with no manufactured urgency.
Matches users to partner bank and NBFC loans. The Compliance agent checks suitability before any recommendation.
Finds coverage gaps, then recommends and facilitates the purchase. Needs an IRDAI corporate-agent or broker licence.
Annual trail on regular-plan AUM as an AMFI-registered distributor (ARN), kept separate from any advice. Regulated by SEBI and AMFI.
Pre-qualified offers based on the user's CIBIL profile, shown only to people likely to be approved.
Phones, appliances, travel and tickets bought through the vault, with every offer stacked. The commission is disclosed on every purchase.
Per-payment fees as a Bharat Connect agent institution, which become meaningful at 100M+ payments a month.
Financial Actions Completed Per User Per Month. An action is a concrete outcome: a bill paid, a score checked, consent granted, savings found, a loan pre-qualified. The Month 12 target is 8, after 4 at Month 6.
From "Hi" to the money picture. The target falls below 2 minutes by Month 12.
Our product–market-fit signal. If users trust us with their data, the product is working.
Across all languages. A misheard financial instruction is a serious failure.
On amounts, scores and balances. Financial accuracy is sacred.
Every player in Indian money today is built around one product or one tap. Agents that act for you are arriving in messaging apps, which proves the interaction model, but none of them speaks Indian money.
Brilliant at the transaction. They know what you paid, not why, and they don't plan what comes next.
Good dashboards for app-first, English-first users. You still do the thinking, one app per product.
Great for affluent card users. Not built for a household running on UPI, chit funds and remittances.
Proactive agents inside WhatsApp and iMessage, with memory and approvals. Muse launched in the US in September 2026. None of them plugs into Account Aggregator, Bharat Connect or India's regulators, or speaks 12 Indian languages.
Trusted, but expensive, conflicted by commission, or only available to the affluent.
Starts from what you want, works out every step in your language inside WhatsApp, with pages that build themselves for each request, and acts on India's rails with consent and a one-word yes.
No app download, and from Phase 2 no smartphone needed, thanks to IVR. Bill payment is the lowest-friction entry point. Once someone is in, the dashboard, credit coach and savings agents do the rest.
A bill is paid, and the confirmation says "Share Sahaya with family". A relative starts with their own bill, then finds the dashboard.
Users get a shareable credit-health card that ends up on WhatsApp status. Scores are social, so friends get curious and check theirs.
One user onboards their parents or spouse by forwarding a WhatsApp message, which unlocks the family dashboard. Managing money across generations is Sahaya's superpower.
Telegram is already live, IVR covers feature phones, SMS is the fallback, and the Android and iOS apps hold the vault. The relationship lives in Sahaya's memory, not in the channel.
Statements are parsed by code before the model sees them, every rupee must come from a tool or arithmetic, and an out-of-process compliance engine can veto any action.
Each regulated activity runs through the right licence or partner: Account Aggregator as an FIU via a regulated partner, AMFI registration for fund distribution, an IRDAI corporate-agent or broker licence for insurance, an RBI-authorised payment aggregator for tokenised cards, and DPDP-compliant consent throughout.
Keys stay on the user's device, cards exist only as network tokens, a secret is released for one approved action at a time, and every release is written to an append-only log.
Every suggestion discloses whether Sahaya earns a commission, suitability is checked before any product, and an Activity log (ask "what did you do?" in any language) shows the user everything Sahaya has done.
Small models handle the specialists and a larger one only routes, documents are parsed without the model, and Sahaya Pro covers the heavy users.
The world needs AI that goes deep in one domain rather than wide and shallow across all of them. For us, that domain is money. Sahaya means companion: a helper that stands beside you.