Realtime AI News
Aina's $399 AI ring puts a screen and joystick on your finger to approve agent actions
Hardware startup Aina has released a $399 AI ring that packs a screen, joystick, microphone and speaker into a wearable for taking notes and directing AI agents. Early-bird preorders opened at $399 before returning to $499, with the company planning to ship in January 2027.
Hardware startup Aina has released an AI ring whose selling point is not step counting but command over your agents. The device is priced at $399 during an early-bird preorder window, rises to $499 afterwards, and is scheduled to ship in January 2027.
It is a chunky ring. A small display on the face shows phone notifications and tells you which app you are currently controlling; a joystick plus gesture support handles navigation; the microphone is for taking notes, firing off prompts to agents, setting reminders and dictating on a Mac; and a built-in speaker lets you take calls on it.
Aina is not a first-time founder's blank slate. Chief executive Apoorv Shankar previously built Mirage, a three-key controller that mapped keys to whatever app was on your Mac screen, so that during a Google Meet call a key could mute the mic, toggle the camera or leave the meeting. The ring carries that idea onto a finger: surfacing the right action for the current context.
Shankar frames the bet as an interface problem. Phones and laptop screens, he argues, were designed for a browsing era in which you first decide which app to open, while proactive agents call for a different kind of control surface. The ring is positioned less as a phone replacement than as a small remote for both AI and non-AI apps.
The category is no longer empty. Pebble Index 01, Sandbar Stream, Oasis and Natura all sell smart rings that push AI interaction toward the hand, and the wider field includes OpenAI's experimental keyboard and Rabbit's attempt to reposition itself as an agent controller. Aina's differentiator is simply more controls: screen, joystick, gestures, mic and speaker in one device.
Early hands-on impressions describe something closer to a smartwatch worn on a finger. It delivers notifications and lets you reply, surfaces meetings, can summon an Uber home, controls music and smart home devices such as Apple TV, though the Apple TV remote itself remains easier to use. Much of the software is still being fleshed out, so where such a device lands in daily life is genuinely unclear.
The product logic is what matters. As agents move from conversation to execution, approving, interrupting and switching tasks become the new interaction bottleneck, and that is precisely what a phone screen handles worst. The questions now are whether January shipping holds, whether developers build for a ring form factor, and whether $499 convinces people who already run several agents from their phones.
Why it matters
Moving agent approval and interruption onto wearable hardware reframes AI gadgets from personal assistants into control consoles for autonomous software.
Nearby Updates
All11/08, 19:05
Qualcomm Demos On-Device MoE Agent at CIIE With StepFun and Partners
At the ninth China International Import Expo, Qualcomm demonstrated an on-device MoE agent built with StepFun, Infris AI and Longsys that handles intent understanding, task orchestration and real-time app control locally. The demo optimizes model, inference, storage and chip platform together to work around the memory and power limits of phones and PCs.
11/08, 17:38
China Mobile and Zhipu Sign Cooperation Deal, GLM Models to Join Mobile Cloud MaaS
China Mobile and Zhipu (Z.ai) have formally signed a cooperation agreement covering domestic compute, large-model services and AI commercialization, Sina Finance reported on November 8. Under the disclosed terms, Zhipu's GLM series models will be connected to China Mobile's Mobile Cloud MaaS platform, spanning compute supply, model services and commercial distribution.
11/08, 16:00
Nous releases fine-tuned Qwen it says strips out the original's state ideology
US open-source AI company Nous has published a fine-tuned version of Qwen that it says removes the national ideological values carried by the original model. The release turns a live debate about whether open weights permit downstream value replacement into a concrete, downloadable artifact.
11/08, 14:54
DeepSeek Nears OpenAI's Cyber Model on Vals Test at About $3 a Run
Vals AI's CyberOffense leaderboard shows an open-weight DeepSeek model finishing close behind OpenAI's cyber model on tests where agents attack simulated networks, at roughly $3 per run. The result narrows a gap that had been assumed to separate restricted frontier systems from freely downloadable ones.