Local AI in your device: How Indians are building hardware to run AI without apps

Summary

How Indian builders are creating useful hardware and devices that can interface with AI without the need for an app, or sometimes even without an internet connection?

At an AI hardware "buildathon", Nehal Joshi, 40, founder of Bengaluru-based education startup Lernok, rediscovered his inner child. His team combined a webcam, computer vision and the chess engine Stockfish to analyse an ordinary chessboard and explain chess positions in English or Hindi, without any expensive digital equipment, and, Joshi claims, without an internet connection.

The chessboard offers a glimpse of AI's move beyond the screen and into the objects around us. A television show had given us this glimpse of this years ago. In the sitcom, Small Wonder, Victoria, aka Vicki, a robot built to resemble a young girl, lived with the Lawson family as their daughter. Her literal-mindedness made comedy out of a familiar problem -- understanding an instruction is easier than understanding the humans giving it.

Better models, cheaper processing and improved sensors are allowing Indian builders to experiment with devices that listen, notice and assist without an app interface. Capital is following that possibility. A16Z (Andreessen Horowitz) has raised a $1.1 billion Machine Age Fund spanning AI infrastructure, robotics and home appliances, while Y Combinator's Requests for Startups includes consumer AI in its Fall 2026 list.

Do we need intelligent devices?

Underneath the new devices built using AI is a familiar desire to have less to keep track of, and reassurance that important things will not go unnoticed. We, as consumers, have always bought appliances to take chores off our hands. These products offer to take some of the noticing function off our minds. But how much access to everyday life will we grant a machine in exchange for its attention?

There is considerable appeal in figuring out how help could arrive without someone first having to ask for it. And the ability to make an "intelligent" object seems to be arriving faster than the actual necessity for such a thing. Sometimes, though, the need is painfully clear.

Ayush Agrawal, 36, co-founder and CEO of the Bengaluru-based, Antler-backed Kubo Care, began building after his father had a stroke when he was home alone and the episode had gone undiscovered for hours. So he made a wall-mounted device that uses radar and machine learning to recognise falls and alert caregivers. Agrawal says that the device is currently deployed across more than 10 communities in India, Japan and the US, primarily through senior living operators. "For an older resident, there is no wearable to remember to charge. Reports on night-time movement have prompted care teams to revise routines," says Agrawal.

The everyday work equivalent of a similarly useful device is one that allows one to listen without also having to take notes simultaneously. Dhananjay Yadav, 34, co-founder and CEO of Bengaluru-based NeoSapien, says their Neo 1 wearable turns conversations into summaries and commitments using speech recognition and language models. The startup raised $2 million in a seed round led by Merak Ventures and reports more than 15,000 customers. Yadav finds that people with frequent, consequential conversations use it more actively than those who bought it out of curiosity.

Others are designing hardware that helps users interact with AI. Sridipto Ghosh, 24, co-founder and CEO of antimattr, based in San Francisco and Bengaluru, is building a ring-and-earphones interface for speaking to AI agents. The startup is part of Y Combinator's Fall 2026 batch. Ghosh reports over 300 paid pre-orders and more than 40 early units sold. "While everyone is building AI agents, we are building the interface," he says. The bet is that people are already talking to AI agents for work, so a device that helps them do so well is useful. On the ither hand, such a device still isn't a need in daily life.

Alysha Maria Lobo, 41, is an early beta user of Plaud's recording devices as well as a NeoSapien user. She is a venture investor in robotics and physical AI who divides her time between Bengaluru and Shanghai. Lobo finds it difficult to imagine returning to her old, non-AI workflow, and sees people accepting even imperfect devices as long as these save on effort.

Privacy and durability

But such a tolerance has limits. An incomplete meeting summary and a missed care alert carry consequences on entirely different scales of magnitude. So does the social negotiations: The person wearing a recorder and the person speaking to them may have different ideas about what the conversation is for, or whether it violates privacy. Yadav says users typically activate Neo 1 for work and put it to sleep afterwards. Raw audio is deleted after transcription, though the transcripts and summaries remain. Kubo Care's radar avoids creating images, although movement data still captures private routines. While these are basic steps when it comes to protecting privacy, there's a degree of constant surveillance that is baked in to the use of such devices.

There is another dependency inside the convenience: is the device you buy repairable, or will it be rendered obsolete when the next software upgrade comes around? Kushal Bhagia, 36, founder and general partner at Bengaluru-based All In Capital, warns against expecting hardware to follow software timelines. Lobo points to gaps in India's component supply, manufacturing and testing infrastructure. Those gaps determine the quality of a device and whether it can be repaired.

And if the company that built the device cannot afford its AI bills, then the smart device may turn dumb fairly soon. "Neo 1 has a one-time cost and no subscription", Yadav says, "but the company bears recurring AI costs." People may buy these objects like appliances while relying on them like services. Owning the hardware does not guarantee that its intelligence will stay.

Building hardware requires upfront spending on components, prototypes, tooling and testing. And all this while demand is still uncertain. AI coding tools, accessible components and investors willing to fund early experiments are making that first step more achievable. "The most exciting AI hardware in India may look ordinary and do impeccable tasks; and what changes is how much of our lives we entrust to it," says Joshi.

Consumer AI is finding a physical presence through hardware, beyond phones and screens, in the spaces we inhabit. As AI moves in, household negotiations will include a new participant. A machine that listens and assists without understanding everything that is being said. We will have to decide when to invite its attention, when to rely on it and when to shut it out. Robot Vicki's domestic misunderstandings had scriptwriters to sort them out. We will have to write our own house rules.

Shalaka Kulkarni (@shalakulkarni) is a Bengaluru-based author who writes at the intersection of culture and technology.