The race to build the next AI interface is crowded with startups. The Sandbar ring, Plaud’s AI pin and desktop notetaker, and Pocket’s credit card-sized pucks are all vying to capture what you say and do. Bee and Friend take the wearable route, while Meta Ray-Bans and Even Realities are betting on smart glasses. Now, a Bengaluru- and San Francisco-based startup, Aina (“mirror” in Hindi), is trying to make its own mark in this crowded field of human-computer interface devices.

The company announced today that it has raised $5.5 million in a round led by Redstart Labs (Infoedge, India) and 360 ONE, with participation from MIXI Global Investments, Antler, and Blume Founders Fund.

The round also drew individual investors, including newly appointed WhatsApp head Kunal Shah, Razorpay co-founders Harshil Mathur and Shashank Kumar, and Scribd founder Tikhon Bernstam.

Aina, previously known as Project Mirage, was founded by Apoorv Shankar, a former VP of Hardware at smart ring maker Ultrahuman. Before that, Shankar ran LazyCo, a hardware interface design startup that made gadgets, including a ring that let users control other devices like a smartphone. Ultrahuman later acquired LazyCo, bringing Shankar in-house before he eventually struck out on his own again.

“I left Ultrahuman last year because I was just super curious about the space of AI interfaces,” Shankar told TechCrunch. “Devices like Rabbit and Humane Pin had launched, and I had my own disappointments with them. However, I was just excited that we are seeing interfaces being a thing now. And as an engineer turned product designer, this was the hottest thing I could imagine myself building.”

The startup’s first product is Dune, a three-key, context-aware “macro” keyboard — essentially a small keypad that runs pre-set shortcuts — that can control the mic and camera in a meeting and run shortcuts or scripts based on the app users are viewing.

Aina developed two other devices: Radiance, a tabletop remote for video calls with a dial for volume and buttons for mic, camera, AI notetaker, voice modulation, and joining the meeting; and Shift, a single-tap “agentic” button — press it once, and it triggers an AI agent to carry out a repeated task — that connects to your phone.

But in early testing, Aina found Dune was the most popular of the three and realized it could bundle features of the other two devices into the keypad. That signal from users is why the company decided to ship Dune first. It wants to learn, in the wild, what kind of tasks users actually want to automate.

Aina said lessons from all three devices will feed into its next product. The company isn’t revealing details of its new device yet, but plans to begin testing with a small group of select users in the coming weeks.

Shankar hinted that the new device won’t be a passive “context capture” gadget — the kind of always-listening ring or Plaud-style meeting notetaker that just records what’s happening around you — but rather a device built to control and invoke agents.

“I think you have enough context, you have in your phone and your laptop all the time, and we haven’t even started using that well. We are building an action-oriented device that will use the context to help you control and trigger workflows,” he said.

As more developers and knowledge workers adopt AI coding tools like Claude Code and OpenAI’s Codex, there has been a steady rise in hardware built specifically to control and trigger those agents. Just this week, OpenAI released a custom keypad for Codex made with Work Louder. Plenty of other options exist too, ranging from keyboard makers to DIY enthusiasts building their own macro controllers.

There are also reports that OpenAI is developing a smart speaker with a built-in AI assistant, and Rabbit R1 has positioned itself as another device for invoking AI agents. Qualcomm, meanwhile, says it’s experimenting with more than 40 devices to interact with AI. With no clear winner yet on form factor — ring, pin, glasses, keypad, or speaker — expect a wave of new hardware bets, and funding rounds, chasing the same question: what does controlling AI actually look like?