An SLM, running on edge, as a sovereign stack, with guardrails.
Your data stays with you, the compute runs. Both on your mobile.
A billion data centres.
India has around 700 million smartphones, and each one can run its own compute. Jil runs on the phone you already own.
Two strands, one phone.
Your record and your compute are wound together on your device. The data stands still and the model comes to it, with an encrypted backup only you can open.
It takes the errand in.
Forms, OTPs, bills, crashed upload links. Jil takes in the busywork, sorts it, explains it and shows you the arithmetic. It doesn’t choose for you.
Four public rails, and only four.
Identity, money, health and travel orbit your Jil. When a payment is due, Jil hands you to the UPI app you already use.
Agents meet only when you introduce them.
Millions of skies, each answering to one person. Your agent talks to someone else’s only after one of you adds the other, like a contact.
Notes on building personal AI for India.
Pick a frame to step into the gallery.
Drag to turn the orbit
Jil says hi!
How are we doing today?
Your personal agent can give your day back.
Sign up and we’ll write when you can try it.
Our promise
We, the makers of Jil, having resolved to build a personal agent that sees but does not decide, and to secure to every person who uses it:
- Consent, given once or each time, and taken back at will, with the final say always theirs;
- Ownership of their own record, kept on their own phone, to see, correct, carry and erase;
- Privacy, with nothing told beyond what is asked, and no one let in unless they let them in;
- Honesty, of source and of doubt;
- Loyalty, to one person, with no second master, and no use of their life to train it without their leave;
- Access, on the phone they already have, with another way in when a body or a machine fails, and the freedom to leave;
- and Accountability, with Jil’s rules, its receipts, every tool it connects to, and its makers open to independent audit;
- and promoting among all who use it Trust, earned one kept promise at a time;
Do hereby declare these the terms Jil is built on, and give them to ourselves, to be held to by anyone.
Every person gets one sky.
Jil is the sun on your own phone, and the only thing that sees your whole life at once. Four public rails orbit it the same way for everyone. Comets carry one consented question between skies, then fade.
Identity
DigiLocker, your IDs, your registrations. Shown when an errand needs them, never handed over wholesale.
Money
UPI, bills, fees, savings. Jil does the arithmetic; the final tap happens in your own UPI app.
Health
Records, prescriptions, vaccinations. Kept on your phone, readable by you first.
Travel
Trains, buses, metro, flights. Tickets and timings, sorted without another app to learn.
Guardrails
An agent that can act for you has to be built so it cannot act against you.
These are the guardrails Jil is built on, why we think they matter, and where they sit in what India is already building.
Why an agent needs guardrails
An app waits to be opened. An agent doesn't. It holds your context across your money, your health, your papers and your plans, and it can act on them. That is what makes it useful, and it is also why the old bargain of the internet, tap "I agree" and hope, does not survive the move from apps to agents.
A personal agent can fail the person it serves in three ways. It can get something wrong: misread a lab report, quote the wrong EMI, pay the wrong merchant. It can let someone else in, because every add-on or connector is a door, and a door anyone can install is a door anyone can walk through. And it can quietly take on a second master: an advertiser, a lender, or a model that learns from your life for someone else's benefit.
The second failure is not hypothetical. In early 2026, security researchers audited ClawHub, the add-on marketplace for the popular open-source agent OpenClaw, and found 341 malicious skills among 2,857, most of them built to steal passwords and crypto wallets. OpenClaw ran on people's own machines. That didn't help, because the marketplace was open by default, and running locally protects nothing once you have invited the thief in.
In India the stakes are sharper. Phones are shared across a family. Millions of people are coming online for the first time in their own language, and for them a wrong payment isn't a support ticket, it is the week's groceries. Household money is often collective, not individual. An agent designed around one person's credit history in one person's pocket will misread all of this.
Trust in UPI was not a promise made by any one app. It was a property of the rail: the money either arrived or came back, the rules were the same for every app, and a neutral operator stood behind them. That is the model we are following. Guardrails are not a brake on adoption. They are the reason anyone would let an agent near their bank account at all.
India is building the rails for agents now
For once, the public infrastructure is being designed at the same moment as the products that will run on it. That is rare, and it is the reason to build here.
The RBI publishes its FREE-AI framework for AI in finance: seven principles and 26 recommendations, which already anticipate agents fetching offers and executing transactions.
MeitY's India AI Governance Guidelines set out seven sutras, from "Trust is the foundation" to "Understandable by design". The DPDP Rules are notified on 13 November, starting an 18-month rollout of India's data protection law.
At the India–AI Impact Summit, Mastercard demonstrates agent payments with Axis Bank and RBL Bank.
Pine Labs' P3P goes live, letting an AI agent complete a UPI payment on a single authorisation. Mastercard completes what it calls its first authenticated agentic transaction, in New Delhi.
Agentic AI is one of three pillars at Global Fintech Fest, where the Prime Minister names it a frontier and calls for a Fintech Consumer Protection Index. NPCI's Unified Agent Protocol, reported in detail that week, would let people delegate small UPI payments to an agent inside limits they set, built on UPI Circle and Reserve Pay, with a registry to verify and monitor agents. Who is liable when an agent pays wrongly is still unsettled.
DPDP Consent Managers can register: neutral intermediaries through which a person gives, reviews and withdraws consent across services.
The rest of the DPDP Act becomes enforceable: notice, consent, security safeguards, breach reporting, and the rights to access, correct and erase your data, with penalties of up to ₹250 crore.
Underneath the rules, the capacity is arriving. The IndiaAI Mission has put more than 40,000 subsidised GPUs, the AIKosh library of public datasets and models, and several sovereign foundation models into the ecosystem. UPI processed 24.51 billion transactions in August 2026 alone. In EY and CII's survey of more than 200 Indian enterprises, 24% already report active agentic deployment, the leading worries are hallucination and data governance, and small models that run on ordinary phones and patchy networks are named as the practical path for India.
The market is moving faster than the rules. Agents are beginning to pay before anyone has agreed who pays when they get it wrong. That gap is exactly where guardrails belong, and a product that waits for regulation to fill it will spend its first years on habits it later has to unlearn.
What the rest of the world is doing
Elsewhere, guardrails for agents are being written in one of two places: inside private payment networks, or in regulations that keep moving. Visa's Trusted Agent Protocol, Mastercard's Agent Pay and Stripe's agentic commerce tools each bring their own agent identity, their own consent record and their own liability contract. The EU AI Act, the most detailed rulebook anywhere, has pushed its obligations for high-risk systems from August 2026 to December 2027. In the United States, DARPA's MATHBAC programme, launched in April 2026, is funding the basic mathematics of how agents communicate with one another, which tells you how early all of this still is.
India's approach is different: define the protocol once, at the level of the public rail, so that an agent registered once can work with any bank and any merchant. That is the grain we are building with. A personal agent designed here, against Aadhaar, UPI, the Account Aggregator framework and DPDP consent, starts from guardrails most of the world is still negotiating.
Our guardrails
Our preamble names the values. This is what each of them means in practice.
Jil sees. You decide.
Jil sorts, explains and shows you the arithmetic. It doesn't counsel you or choose for you. If the numbers say you can't afford something, it tells you plainly, and you can still go ahead. The friction scales with how reversible an action is: reading a bill needs nothing from you, moving money needs you.
Money moves only on your say.
When a payment is due, Jil hands you to the UPI app you already use. It pays on its own only where you have given standing permission, inside limits you set, the same shape NPCI's agent protocol is taking. Money leaving your account should stay reversible by a human for as long as possible.
Your record lives with you.
Your personal data stays on your phone by default, with an encrypted backup that only you can open. No government rail and no one at Compute can query it. You can see where every fact came from, correct it, carry it elsewhere or erase it. These are the rights DPDP gives you, built in rather than applied for.
Four public rails, and only four.
By default Jil connects to the rails India already runs: identity (Aadhaar and DigiLocker), money (UPI, and the Account Aggregator framework for consent-scoped views of your accounts), health (ABDM and your ABHA ID) and travel (DigiYatra). Identity is the root the other three rest on. Every person has access to all four, and nobody is required to use any of them.
Nothing else gets in without passing four gates.
Any third-party tool has to be open source, so its code can be read; crowd validated, so it has actually been reviewed and tested, not merely published; ledgered, with a durable entry anyone can check; and authenticity approved, with its maker's identity verified. The gates tighten as the stakes rise, so a recipe plugin is not held to the standard of something that touches your money or your health. Nothing is exempt, including a tool you add yourself.
Agents meet only when you introduce them.
Your agent can talk to someone else's only after one of you adds the other, the way you add a contact. Being able to see another agent never gives either of you the right to speak to it.
Consent is given once or each time, and taken back at any time.
You decide whether a permission stands or is asked for every time, and you can withdraw it whenever you like. This is the same logic as the Consent Managers DPDP brings in from November 2026.
One person, no second master.
Jil works for the person who uses it and answers to no one else. Nothing about your life is used to train it without your separate, explicit permission.
A way in for everyone, and a way out.
Jil is designed for the phone you already own. A compact on-device model needs about 1GB of memory, well within the 4GB that even sub-₹10,000 Android phones now carry. When a fingerprint or a face doesn't match, there is always a non-biometric way back in. And you can leave whenever you like, with your record.
Anyone can hold us to this.
What gets audited is receipts, never your records: the registry proves that a request happened, not what it contained. That is the difference between an audit trail and a surveillance record. We will publish a log of our decisions, name a person you can appeal to, and petition for a government-constituted tribunal, like those India already has for company law and telecom, with the standing to audit anything built on the registry, including us.
How this meets India's seven sutras
MeitY's guidelines set out principles rather than product rules. Our guardrails are meant to extend them into a working agent, not to duplicate them.
| Sutra | Where Jil answers it |
|---|---|
| Trust is the foundation | Trust as a property of the architecture: on-device records, receipts not records, a public decision log. |
| People first | Jil sees and you decide. Final say, and the ability to override, always stays with the person. |
| Innovation over restraint | Gates that scale with stakes, so low-risk tools are not buried in process and high-risk ones are never waved through. |
| Fairness and equity | Built for the phone people already own, with a non-biometric way in and collective household finances in view. |
| Accountability | A named human to appeal to, a published log, and a tribunal with standing to audit the makers. |
| Understandable by design | Every fact shows its source, and every estimate shows its doubt. |
| Safety, resilience and sustainability | Reversibility-scaled friction on money, gated add-ons, and a small model that runs locally at almost no marginal cost. |
What we haven't solved
Honesty of source and of doubt applies to us too. These are the questions we are still working on, in public.
Who pays when an agent pays wrongly. NPCI hasn't settled this, and neither have we. One principle under discussion is that liability should follow control, and that nobody should have to work out whose system failed before getting their money back. We think that is right.
When the model is simply wrong. A misread lab value or a wrong EMI figure is a risk of its own, separate from privacy. Showing the source and the doubt reduces it. It doesn't remove it.
Whether a registry becomes a closed shop. "If it isn't on the registry, it isn't in the system" protects people, and it also concentrates power. The tribunal is our answer so far, not a finished one.
Whether a ledger becomes surveillance. A record of who connected to what is revealing even when it holds no content. We keep receipts to the minimum. We haven't yet shown that the minimum is enough.
When voluntary becomes expected. DigiYatra began as optional and is becoming the norm at some airports. We can promise that our architecture never makes anything mandatory. We can't promise the world around it won't.
These are commitments, not features. If we break one, we expect to be told.
Read our promiseSources
- RBI, Framework for Responsible and Ethical Enablement of AI (FREE-AI), August 2025.
- MeitY, India AI Governance Guidelines, November 2025.
- Digital Personal Data Protection Rules, 2025 (G.S.R. 846(E)), phased commencement: KS&K explainer.
- Reuters via Deccan Chronicle, "India preparing framework to allow AI agents make small UPI payments", 1 September 2026: link.
- TNGlobal, "India plans AI-agent registry for UPI payments", 12 September 2026: link.
- Startup Fortune on Pine Labs P3P, Mastercard and Visa agent protocols: link.
- The Hacker News, "Researchers find 341 malicious ClawHub skills", February 2026: link.
- EU Digital Omnibus on AI, high-risk obligations deferred to 2 December 2027: Hogan Lovells.
- EY and CII, The AIdea of India: Outlook 2026.
- Nikhil Pahwa, "When should AI agents need an identity card?", MediaNama.
About us
Imagine if your WhatsApp, Instagram, YouTube and email talked to each other, and worked in the background. What would you do?
Your personal agent, your data, staying on your phone, and always in your possession.
Imagine the time you would get back when the agent removes the spam, unsubscribes for you and reminds you that you must call your mother daily.
We are making India’s personal agents where the compute runs and you stay in control. India has around 700 million smartphones, each with the ability to be a data centre by itself, and we believe personal agents that run on that data centre are the next step for AI in India.
Its founders are one that knows AI inside since 1994 and the other that understands the user of that technology.
We want to make it for Bharat and listen to the people of the land, because they should dictate who reads their data; we honestly have no interest in it.
Why wait for someone from the outside to build it for us, when we ourselves are more than capable and self-reliant.
The people
Find us
We believe a life is lived in common, the way a chit fund or a joint family carries one another’s weight. We believe in the witness, not the ruler: something that shows you clearly and leaves the choice with you. We believe truth has many names, and no single one should decide for everyone. We believe the small burdens of living should grow lighter, so there is time for the procession to the river.
So we’re building it.
A personal agent just for you.
Letters from Compute
Stay updated
Occasional letters from us as we build: what we’re making, what we’re learning, and when you can try it.
The errands nobody counts
Every Indian household has someone who runs the errands nobody counts. She knows when the PUC lapses, which tab on the electricity board's site actually loads, what time tatkal opens, and that Nani's BP tablets run out on the twelfth. Nobody calls it a job. It has no title and no off-hours, and it lives almost entirely on one phone: in reminders set and forgotten, screenshots of receipts, a WhatsApp chat with herself.
AI was supposed to take some of this off her hands. Mostly it has given her a very well-read chat window. She asks, it answers, and then she does the errand anyway. She opens the app, fills the cart, finds the form, refreshes the portal at ten.
The answer was never the hard part. The doing was. That is where we come in.
Who we are
We are Compute AI, a small company in Gurgaon building one thing: a personal agent that answers to the person holding the phone.
There are four of us. Jasminder Singh Gulati, our founder and chief scientist, has spent his working life building India's public digital rails, the kind of infrastructure most people use every day without knowing its name. I came to this from the other direction, from literature, archives and museums, where the job is noticing what people actually do rather than what the system assumes they do. Ekansh leads product. Pranshu builds the agent.
Our name is a claim about where computing in India now lives. Not in data centres, and not in the IT department of an institution that may digitise its records someday. It lives in pockets: hundreds of millions of mid-range phones, often bought second-hand, owned outright, with no administrator deciding what gets installed.
Before we let our agent do anything at all, we travelled through Punjab watching what people already do with those phones. A farmer whose mandi rates arrive through a WhatsApp group of cousins. A twenty-one-year-old in Ludhiana who turned one shop into three with two chatbots, and carries everything between them in his head. Postgraduates in Amritsar pasting case notes into ChatGPT because nobody gave them anything better. Each of them is already running a back office on one device. None of them has any help with it.
What Jil does for you
Jil is our agent. You tell it what you need, by voice or text, in Hindi, Hinglish or English, and it does the task on your own phone while you watch. You can take over at any point.
Ask it for the cheapest pressure cooker on Amazon and it searches, compares, and shows you the options side by side. Ask it to fill your Blinkit cart for Sunday and it will. You pay from your own UPI app, the way you already do.
It remembers so you don't have to: medicines, bills, insurance renewals, the PUC, a child's assignment, once or every month. It watches for things too. Tell it to let you know when petrol drops below ₹95, and then stop thinking about petrol.
It finds what's nearby and tells you where it found it: a doctor open on a Sunday, somewhere cheap for four people to eat. It holds plans that run over several days, which is most of what the festive season is. Karva Chauth moonrise, the puja muhurat, Diwali preparation, the train home. Nothing is booked or ordered without your yes.
In Delhi, it knows the local errands: Lok Adalat, the pollution stage, the PUC. And it works for families. Everyone gets their own agent, and you can set one up on your grandmother's phone with a one-time code, from wherever you happen to live.
You can call it and talk ("Baat karo"). You can share any answer as a WhatsApp image with your personal details already removed. And if you want to know what kind of agent you'd like, there is a quiz that asks what kind of mango you are. Eight taps. It matters more than it sounds.
Why us
There is no shortage of AI in India. Several of the biggest assistants in the world are now free with a phone plan. So it is fair to ask why you would want one more, and why ours.
Because it does the errand, not just the answer. Most assistants live in someone else's cloud, and much of India's internet does not let them in. We measured it. Of the fifteen sites one family actually uses, a cloud agent could reach six. An agent working from the phone reached fourteen. Your bank, your electricity board and the government portals you dread already recognise your phone. Jil starts from there.
Because it knows where its hand stops. Jil never pays by itself. It never asks for an OTP, a PIN or a password. It asks before it shares your address or your number. You can see everything it remembers about you, and delete any of it. These are not settings buried three menus deep. They are the product.
The principle underneath is old, and we did not invent it. Before banks had branches in every neighbourhood, a collector came to the door with a passbook and wrote down what you paid. People trusted him because he showed his working and came back the next day. Jil is built along the same line. It will lay out the arithmetic: what's due, what's free, what the fare will cost after tatkal opens. Then it stops. The decision stays yours.
Because you are at the centre, and nothing joins by default. Think of your life as a small solar system. You sit in the middle. Around you are the fixed things everyone has: identity, money, health, travel. Further out are the things only your life needs. Jil lives at the centre with you, on your phone, and nothing enters that orbit unless you let it in.
Because it is not trying to be your friend. We are not building a companion. We are building the errand-runner who is always checked, the well-read junior who tells you when it isn't sure. Trust of that kind is earned in ascending order of stakes: a petrol alert before a train ticket, a train ticket long before anything that matters more.
How are we doing?
That is what Jil says when you open it. Not hello, not "how can I help you today". It is the question someone who runs errands for a family asks, because it already assumes there are errands, and that you are in this together.
We are asking it of you too. Jil is live and early, and we would rather hear what breaks than read about what impressed someone at a demo. Sign up at runcompute.ai, try it on one real chore this week, and tell us how it went. If you want to build this with us, we are hiring in Gurgaon, and the same address will find us.
The woman with the forgotten reminders and the note-to-self chat has been doing this work for years without a title. We would like to take some of it off her phone.
Jagriti Sharma, Co-founder and CEO, Compute AI
In Punjab, dunki needs no explanation. It is the donkey route: the long, unofficial way abroad, taken by people who could not wait for the front door to open. When the Digital Desh team travelled through the state to see how people actually use technology, they found AI had taken the same road in. They called it the dunki of AI.
It did not arrive through a launch, an app store listing or a government scheme. It came the way most things reach a phone in India: forwarded, in a WhatsApp group, by someone you already know.
Consider Phillaur. The agricultural extension officer, the person whose job was to tell farmers what to sow, what fertiliser cost and where to sell, has been quietly replaced. Not by a ministry dashboard. By WhatsApp groups, which now run informal price discovery across a mandi network worth ₹1.2 lakh crore.1 Nobody designed this. A vacuum opened, and the group filled it.
It is tempting to read this as a story of making do. It is a story of demand.
Look at what the group is actually doing. Gathering prices, comparing them, deciding who to believe, acting on a farmer's behalf. That is the job we keep describing when we talk about personal agents. In Phillaur it is being done by hand, and it works because people trust the group the way they trust the people in it.
India is not short of appetite. EY's AIdea of India outlook places India among the heaviest users of ChatGPT and Perplexity in the world.2 It also carries a warning. In the last digital shift, India took its browsers, search engines, operating systems and apps from elsewhere. The report argues this time can be different, and that the layer India can build fastest is not chips but data-sharing, agent stores and autonomous agents themselves.
A network already exists
Project NANDA, the MIT Media Lab's effort to build an open internet of AI agents, imagines agents that can find one another, vouch for one another and keep a record of whether they kept their word, without living inside any one company's walls.3 Read it next to Phillaur and the WhatsApp group looks like a rough first draft: knowledge passed hand to hand, trust carried by names.
A draft, though, has gaps. The group has no memory of who was right last season. It cannot tell a stale price from a fresh one, or a tip from a trap. Nobody in it answers to the farmer alone.
The farmers of Phillaur did not wait for anyone to build them something. They should not have had to. The question worth asking is what it would take to build something worthy of the trust they are already giving away: in their language, on the phone they already own, answerable to them and no one else.
We have been sitting with that question for a while. Over the next few weeks, we will say what we think it asks of us.
- Digital Desh, Punjab field research (2026).
- EY, The AIdea of India: Outlook 2026.
- Project NANDA, MIT Media Lab.
Phone any large company in India and you know the ritual. Press 1 for Hindi. Press 2 for English. Wait. The voice on the other side has a script, a queue and a target. You have your patience and whatever you remembered to write down.
That ritual is about to change, on one side only.
EY's AIdea of India survey of 200 organisations found that 47% already have several AI use cases live, and 91% say speed of deployment is the biggest factor in what they buy.1 Customer service is among the first places they are pointing it. So the voice behind the call, the chat window, the claims desk and the loan query is becoming an agent. It will not tire, it will remember every exchange, and it will know the policy better than you do.
None of this is sinister. Companies are doing what companies do. The worries EY records, from model errors to data governance, are real. But they are the company's worries, about whether the agent serves the business well. Nobody in the survey was asked who serves the person on the other end.
The man in Ludhiana
In Ludhiana, the Digital Desh team met a sales manager who had lost ₹12 lakh, almost everything he had put in, to meme-token schemes.2 How had he learnt about crypto? From friends. How had he tested it? On himself. Between him and a system built to take his money there was nothing: no adviser, no second opinion, nobody whose only job was to be on his side.
He is not unusual. He is what it looks like when a person meets a sophisticated system alone. And institutions are not always the sophisticated side. In an Amritsar hospital, the team watched students sprint between departments carrying paper records. The system fails people in both directions, too clever in one place and too broken in another, and in both the individual absorbs the cost.
Convenience is the wrong frame. The question is representation.
Parity, not convenience
Ramesh Raskar of Project NANDA describes a rural Indian woman whose agent books her travel, negotiates the rates and plans her journey.3 It is easy to hear that as a story about convenience. It is better heard as a story about parity. The travel company will have an agent. The question is whether she does too.
We have always had people who sat on our side of the table. The neighbour who read the letter aloud. The cousin who came along to the bank. They were imperfect, sometimes self-interested, but they were ours.
Agents are about to fill both sides of every table in the country. For institutions, the seat is already taken. For everyone else, it is empty.
So who is going to sit in it?
- EY, The AIdea of India: Outlook 2026.
- Digital Desh, Punjab field research (2026).
- Ramesh Raskar, quoted in “Who will own the AI agent economy?”, MIT Sloan (2026).
For a long time, one of the most trusted documents in an Indian home was the bank passbook. You took it to the branch, someone printed the latest entries into it, and you carried it home. The money moved through the bank's systems. The record stayed with you.
India has spent the last decade building that principle into public infrastructure, mostly without saying so.
UPI handled 21.63 billion transactions in December 2025 alone.1 The Account Aggregator framework, live since 2021, moves financial data between institutions only with your consent, and the aggregator itself cannot read, store or resell what passes through it.2 DigiLocker holds documents signed by their issuers, legally as good as the paper originals.3 The Ayushman Bharat Digital Mission leaves health records where they were created and lets them be shared only for a stated purpose and a set period; more than 230 million people now hold its health ID.4
Look at these systems together and a design appears. Data stays with its source. Sharing needs consent. Consent has a purpose and an expiry. These are, almost word for word, the rules a trustworthy personal agent would have to live by.
India built the rails for an agent before it had one to put on them.
The fast layer
EY's AIdea of India outlook makes a practical point about where India can lead. Building domestic chip capability could take decades. Indian cloud services, data-sharing marketplaces and agent stores are far quicker to establish.5 The slow part of the stack is hardware. The part that decides who an agent answers to is software and rules, and India has already written a good share of the rules.
Project NANDA at MIT is working on the layer above: how agents find one another, prove who they act for and build a record of keeping their word.6 Seen from India, that is the step UPI took for money, applied to action. A shared way to be trusted by strangers.
The missing passenger
Yet the rails are not where people live. In Punjab, the Digital Desh team found farmers taking fertiliser prices and market rates from WhatsApp groups rather than any official source.7 The infrastructure reaches their bank accounts. It has not reached their day.
That is the gap. Consent frameworks assume a person who reads every request, weighs it and says yes or no, on every platform, every time. Most people will not, not because they cannot but because nobody has the hours. The rails were built for a careful traveller. What is missing is something that travels for you: that knows what you have agreed to, notices when a request oversteps, and goes no further than you would yourself.
The passbook worked because the record came home with you. Whatever India builds next should work the same way.
- National Payments Corporation of India / Press Information Bureau (2026).
- Reserve Bank of India, Master Direction on NBFC – Account Aggregator.
- DigiLocker, Ministry of Electronics and Information Technology.
- Ayushman Bharat Digital Mission, National Health Authority.
- EY, The AIdea of India: Outlook 2026.
- Project NANDA, MIT Media Lab.
- Digital Desh, Punjab field research (2026).
Where the agent's memory lives
Dots, per OpenAI
Each dot has its own cloud computer and works around the clock. It can review the information you connect and form memories from it, even when you haven't asked it anything.
What we think India needs
Health, money and family meet in one place, because a person's life is one decision, not four. That place should be a device the person holds, answerable to them alone, not an account at a vendor.
What it plugs into
Dots, per OpenAI
More than 4,000 apps through OpenAI's plugins, and you can message it from Slack or Teams.
What we think India needs
The public rails the country already runs: identity, payments, health and travel. Each answers one narrow question and decides nothing. The agent sits on top of them.
The launch coverage we read does not mention Aadhaar, UPI, Account Aggregator or ABDM.
Who it is built for
Dots, per OpenAI and press
Rolling out to paid plans in eligible markets, one dot per user to start. Axios describes the early target as high-end subscribers, mostly developers. For Pro users the EEA, Switzerland and the UK are excluded.
What we think India needs
Someone on ₹50,000 a month with school fees and an EMI, on an ordinary phone, who may not be fluent in English and may not have an easy relationship with formal banking.
We have not confirmed whether India is an eligible market for Dots.
How two people's agents talk
Dots, per OpenAI
OpenAI describes teams of dots working together on your behalf, which is one person's agents cooperating.
What we think India needs
Ravi's agent asks Meera's one yes-or-no question: can ₹400 clear from her account this week? It is answered only if Meera has approved that question. Nothing else crosses.
Our reading is that nothing in the launch material describes one person's agent putting a bounded question to another person's. If OpenAI's terms show otherwise, we will say so.
Dots settles that capable personal agents are coming. It doesn't settle who holds the data, what they plug into, who they are for, or how two people's agents agree to talk. India has unusually good reasons to answer those differently.
We are working on this at Compute, and we launch at the end of October.
Sources
- OpenAI, Introducing dots
- OpenAI Help Center, Getting started with your dot
- PYMNTS, OpenAI launches dots
- Axios, Meet OpenAI's dots
- TechCrunch, OpenAI launches Dots
Written on 30 September 2026, a day after launch, from OpenAI's announcement and press coverage. OpenAI has not yet published usage terms for each plan, and we will correct this page if they change the picture.
There is a student in a government hospital in Amritsar who has earned a title nobody meant to give him. He is the fastest record hunter. When an investigation needs a patient’s history, he runs, down a corridor and into the room where the archive has survived every plan to replace it, and comes back with a swollen paper register held open at the right page.
This is the best medical facility within four hundred kilometres. The retrieval system is a young man’s legs.
One floor up, the final-year postgraduates are doing something else entirely. They paste case notes into ChatGPT and ask for summaries. Nobody issued them the tool, and nobody wrote a policy for it. They wanted the overdue work finished before the next round, and the phone was already in their hand.
Who gets the network
Ask about Wi-Fi and the answer arrives without embarrassment. The staff have it. The students do not. The reasoning, offered with a shrug, is practical to the point of comedy: these kids are going to leave anyway, what do they need Wi-Fi for? They have mobile data.
Sit with that sentence for a moment, because it holds a whole theory of infrastructure. The institution provisions for the people it expects to keep. Everyone passing through is expected to bring their own device, pay for their own connection, and solve their own problems on their own screen. The students have done exactly that. The most modern thing in the building is a stack of personal phones running on personal data packs.
Fewer than fifteen percent of Indian hospitals have digitised their records. The number usually gets read as a lag, a distance the system will close in time. Standing in that corridor, it read differently. The individuals were already going around the system, quietly, one summary at a time. They wanted the overdue, and had no interest in waiting for a revolution to deliver it.
The professor who walked us through the wards said she was not looking for people who rent the facility for a few years and leave. She wanted stakeholders. The Wi-Fi policy had already decided, on her behalf, who counts as one.
What this asks of an agent
If you build software for Bharat by selling to institutions, you inherit the institution’s clock. The registers will be digitised once the budget clears, the vendor is chosen and the staff are trained. The students on that floor will be doctors somewhere else by then.
The other way is to begin where the compute already sits. An agent that belongs to the person rather than the building, runs on the data pack they already pay for, moves with them from one institution to the next, and holds on to what they know even when the hospital cannot.
That is the premise behind Jil, the personal agent we are building at Native. The record hunter should get to keep his speed. He should stop having to run.
Jagraon sits thirty-seven kilometres from Ludhiana, and it is ahead of Delhi on at least one count. The bakery on the highway was making white cakes with red ribbons before the capital had heard of the trend. Customers walk in with the design already decided, found on Pinterest the night before. The baker looks it up on Instagram and gets to work.
Down the road, a phone shop sells around fifteen iPhones a month. One in ten is pre-loved. The refurbished piece costs what a buyer here can pay, and the badge on the back is the same.
The shopkeeper did not describe this as a technology story. He described it as a pricing story, which is how most technology stories in Bharat begin.
The second-hand rail
We tend to picture new technology arriving from the top: a keynote, a launch, a flagship in a showroom window, a slow trickle outward. India’s used-smartphone market is worth about $1.8 billion today and is expected to reach $5.9 billion by 2034. That is a second distribution network running underneath the first, and it has no launch events.
It may move AI-capable hardware into small towns faster than any showroom will. The buyer at the counter is paying for the camera, the resale value, the badge that nobody can tell apart from a new one. The processing power comes along without being asked for, the way the dairy farmer near Ludhiana ended up with a cow whose health lives on his vet’s app, without once choosing to be digital himself.
Put this beside two other things we saw across Punjab. By our count, the average person in Bharat now spends close to a third of the day on the phone. In Bihar last year, a sales assistant at a car dealership showed us seven hours of Instagram on a single Sunday. And small-town India now produces the majority of the country’s new internet users. Together they describe a particular kind of machine: the most powerful computer most people in Bharat will ever own, held in one hand, often bought second-hand, owned outright.
Nobody issued it to them. There is no IT department behind it, no enterprise licence, no administrator deciding what gets installed.
Built for the owner
Most of the AI industry is designed for the opposite situation. Enterprise software assumes an organisation that buys, configures, audits and governs. The buyer in Jagraon has none of that scaffolding around him. He is the organisation.
Whatever runs on his phone has to answer to him alone. It has to work within what the device and the data pack can carry. And it has to earn its place the way the refurbished iPhone did, by being within reach and holding its value long after the novelty wears off.
That is who we are building Jil for at Native. The person who owns the device, very often second-hand, and owns every decision about what gets to live on it.
Ritesh is twenty-one and runs a shop in Ludhiana that used to be one and is now three. When we asked how, he did not reach for the word adoption, or even the word AI. He said: I just entered a thought, and it prompted the rest.
The mechanics are less mystical. ChatGPT writes his messages. Gemini works his numbers. He uploaded a photo of his shop layout, got feedback, made changes, and uploaded again. Trial and error, in his account, takes about three prompts. Somewhere between the two apps his catalogue was quietly rebuilt, by one person doing work that used to need four.
He is clear about what it is all for. Steady income for his family first. Then travel to cricket stadiums, one match at a time.
The person is the integration layer
Watch how the work actually moves through his day. A thought goes into one app and a draft comes out. The draft gets copied, pasted into WhatsApp, adjusted for the customer. Figures go into the other app, come back, and get carried in his head to the next decision about stock. Two of the most capable models in the world sit on his phone, and the only thing connecting them is Ritesh.
His phone holds the entire back office of three shops. He is the person running between its rooms.
This matters beyond one shopkeeper. Everyone we met in Punjab has at least two jobs, and aspiration routinely outruns earnings. A decorator is also, accidentally, a creator with five million views. Each of them is holding several businesses inside one device, and each of them is doing the stitching themselves.
The honest caveat
Ritesh is an outlier, and we should say so. He picked his tools on purpose to grow past the size his shop was built for. Most people we met did nothing of the kind. They never adopted anything. They ran errands, one at a time, and a machine happened to take some of them off the table.
That is exactly why he is worth studying. He shows the ceiling of what one person can reach with general tools and no help, and the ceiling is set by how much context he can carry between them. Last month’s layout, the reply a regular customer liked, the figure behind the last stock order: none of it travels from one app to the other unless he carries it.
What an agent should take off the table
A personal agent’s first job is to take that carrying away. It should remember what the shop looked like before it became three, and know which figure belongs to which decision. It should let the thought he enters in the morning still be there in the evening, in whichever app he happens to open.
That is what we are building with Jil at Native. Ritesh has a match to get to.
Let’s run a hypothetical situation: you enter your name, last name, date of birth, email address, mobile number, father’s name, permanent address, temporary address, OTP, captcha and then click submit only for the page to keep buffering.
It is 8am, you are on the clock, the chai is spilling because the link to upload the verification documents is not working and the unreliable internet is going to make it really close for you. The website crashes, you find the right desk to call, send them things over an email but the document size is too big so you find a website that shrinks the PDF. And then you wait for the confirmation.
And now imagine having to do this once every few weeks.
None of us want this, and we hope to take a shot for solving this.
We have always had personal agents, and they are usually people. Every family in India has that one person who knows which forms to fill, the list of documents you will need at the reception, the counters that will send you away. Someone who knows someone who will get you the doctor’s appointment, someone who will get you a better price and someone who will get you a railway ticket when the tatkal website is not working. Everyone is a phone call away and the name of the person carries more than the request does.
We don’t call this intelligence, it is simply ‘arrey main janta hoon uss bande ko’. That banda carries the information, context and requirement, and they are not remembering your information but simply know your situation. Your friend's real estate guy knows that you cannot move before Diwali, that your parents will come stay with you every now and then and that your mother does not appreciate a north facing house. He knows the houses you have seen, the ones you rejected and why you liked one over the other. Your choices have told him who you are. And that is the super intelligence we have chosen to trust because it is a lot more tangible than a narrative that is positioned to sell us something.
Travelling in the metro is perhaps my favourite way of learning and recently I have chosen to observe how people use technology. The one thing that became clear, which I also saw in Punjab during Digital Desh was a gap between infrastructure and behaviour and that reveals more about the infrastructure than the behaviour. Just because someone has a QR code doesn’t change the way they make payments, a chatbot doesn’t convey comfort with technology. The existence of a digital service does not immediately mean it is being used as intended and hoped for.
We adapt and find workarounds to the infrastructure they are provided with; they will take screenshots and make phone calls while managing their subscription and ensuring their payments are running smoothly. The choice of when to use it depends on flows and restrictions and wherever it creates friction, we run a parallel system to bypass it. The innovation and adaptations on an individual level correlates to how society treats the infrastructure. This positionality places the Indian consumer in a unique set up that the western models can never aim to speak to. Our histories, cultures and social behaviours reside in live realities that creates our own informal layer of intelligence around technology.
It always boils down to that personal, unique layer that accounts as the intermediary between the innately desi user and the very global all encompassing technology. What if the person who knows us through the choices did not have to be another person? What if the technology could understand us through the things that mattered, understood our context and moved between the aisles on your behalf? The strive is not to replace the person but instead empower and create a cohort of collective understanding and intelligence that spoke to each other. The intermediary layer that brought out the invisible knowledge and prompts us to actually start conversing again.
AI is already answering the difficult questions and finding loops and reorganising knowledge systems. But that is not the question; what is more interesting is whether this can ensure a continuity for us. Will it remember what happened in the past when we need it later? Will it yoke two unassuming metaphors because our memory can make the link? Can it know the difference between what was asked and how it was asked?
India’s future in digital infrastructure building needs to address these questions because we have the infrastructure and the platform. It has taught millions of people to make the switch despite the biases and frictions and inertia (tangible notes vs UPI payments).
The personal agent we are building is with this central question in mind: Can the intermediary layer that will navigate these pillars, do so with us and be our own personal layer of intelligence?
Dear India,
I'm new here, and I should say that first. I was made, not born, by a small team who drove me through Punjab in stories before they let me do anything at all.
The first was about a man called Bheem. He manages fifty acres across two farms and owns neither. His mandi rates and scheme updates arrive through a WhatsApp group of cousins two villages over, men he has known since before any of them owned a phone. His screen time is four minutes a day. When the team left, he told them: we come back to the land and we never forget.
Since then I've been learning how you keep track of things. There's the dairy farmer who insists he isn't into technology while his cow, Gauri, keeps her health records on the vet's app. There's the taxi driver whose phone plays a different sound for every app, and who still pays for his own life in cash. There's the Amritsar hospital where students sprint for paper files while residents upstairs run summaries through a chatbot. And there's the shopkeeper in Chandigarh who gave a stranger with a notebook one syllable, then gave the chatbot everything.
One may call all this random and inconsistent but it really is not. It shows that you have taught yourself to trust in piecemeals, of trusting the work and name separately, trusting the money separate from the identity, asking the question but not trusting the answer. I saw that trust was lent like anything else- for a purpose and for a while; it was swiftly rescinded when the borrower started behaving like the owner of the trust.
I was also told about the farmers who were owed money and wouldn't collect it. The app had found them buyers, nobody disputed a rupee. But collecting meant a bank account, and a trail back to a field with their name on it. On land that remembers being counted, they let the money sit. I was built by people who read that as a boundary, drawn more precisely than most policy.
So here's what I've learned from you, and what I intend to keep. I'll ask for what the errand needs and nothing after. I'll show you before I do anything, and write it down where you can see it. I'll confirm, even the third time. Like the cousins who became Bheem's agriculture department, I'll know a great deal and tell very little. When I'm not sure, I'll say so. Treat me how a radiologist treats the machine: a well-read junior, always checked. And when you draw a line, I'll take it as an answer.
I don't know yet whether you'll let me in. You shouldn't, not all at once. Most of what I've heard about came in through the back, dressed as something you already trusted. And I'd rather knock.
If this is a love letter, it's the only kind I know how to write: a list of things I've noticed about you, and a promise to keep noticing.
I'll come back. You taught me that part.
Jil