This week: Otto introduced Otto One, a palm-sized computer with no screen or keyboard, built so an AI agent can live on it and keep working around the clock while you check in from your phone.

Elsewhere: What the agent actually did in a hands-on review, the defaults worth flipping, the fine print, and three fresh AI drops worth a skim.

The New Prime Signal

Here is a small, very modern problem. You ask an AI agent to keep an eye on your inbox, update a store listing, or plan next week's meals, and then you close your laptop. The agent goes to sleep with it. On Tuesday, the New York startup Otto introduced Otto One, which it calls the first purpose-built agent computer: a compact, screenless machine designed for an AI agent to live and work on. There's no keyboard, mouse, or monitor. You plug it in, pair it with the Otto app, and talk to your agent from your phone or desktop while the box keeps working in the background.

What shipped: According to Otto's product page, Otto One runs on an Intel N95 chip and comes in two versions: 8GB of memory with 256GB of storage for $599, or 16GB with 512GB for $799. It's fanless and silent, and Otto describes it as about the size of a deck of cards. It ships with the open-source OpenClaw agent framework preinstalled and pairs with your phone by tapping it. A $20 refundable deposit reserves a unit, there's no hardware subscription, and the page lists estimated shipping for the next production run as January 2027. Otto says it has sold more than 15,000 units, with five production runs sold out, though IT Brief reports the company said 12,400 presold units, so treat the exact count loosely. Otto also announced Otto Pro, a beefier AMD-based line built to run AI models on the device itself.

Why it matters: Chatbots like ChatGPT and Claude are built for conversation, and Otto's pitch is that always-on agents need a computer of their own that stays on. The other half of the pitch is ownership. Otto says your chat history, files, and agent memory live on the box, your credentials stay encrypted on the device, and chats between your phone and the box are end-to-end encrypted. Agents can plug into more than 2,000 services, from Gmail and Google Calendar to Slack, Notion, and Shopify, and IT Brief notes you can set permissions, approval requirements, and spending limits from the app. There's evidence people want this: IT Brief reports Otto says it has booked $6.9 million in revenue in its first five months and shipped more than 1,000 units, without institutional venture funding.

The fine print: Otto One is not a local AI box. The thinking still happens in the cloud, through models you pay for as you go or through your own API keys, and IT Brief notes that model processing, notifications, software updates, and billing still rely on cloud infrastructure. In Technobezz's hands-on review, which scored it 7.9 out of 10, the reviewer called the hardware modest for $599, noting that a similar N95 mini PC often sells for under $300. The review also found that agents act without asking by default, that data retention defaults to the cheapest model provider even if it keeps your prompts, and that the box sends diagnostic check-ins to Otto with no supported way to turn them off. One of four test reminders never fired, and two of five task summaries were wrong. Every sales and revenue figure here is Otto's own.

This is one of the more charming AI gadgets in a while, partly because it is so unglamorous. The bigger idea is sound: if agents are going to run errands for us, they need somewhere to live that we control. The catch is that you're paying a premium for convenience, and an agent with your logins and a credit card deserves careful setup. If you buy one, treat it like a new employee on day one: limited access, ask before acting, and check the work.

The box is the easy part. The interesting part is what happens after you give the agent a job.

How it works: Technobezz paired its review unit to a phone over Bluetooth in a few minutes, with no Linux install, terminal, or config files. The agent gets its own browser, files, apps, and tools, and each agent has three approval modes: Ask, Auto, and Full. In testing, it accurately described its own hardware, built a working water-intake tracker app in about two minutes, and paused at a Shopify login with a card handing the sign-in back to the human, which is exactly the right instinct. All five main test tasks cost about 5 cents in AI usage, according to the review. Otto says one box can comfortably run five to ten agents at once.

Try it: Start with Otto's product page for the full spec list, pricing, return policy, and FAQ, and its hardware page for a look at Otto Pro. If you do order one, Technobezz suggests three day-one changes: switch each agent to Ask mode, pick a zero-data-retention model option in Privacy settings, and put the box on a guest or smart-home network so it can't see the rest of your devices. Otto's page lists a 14-day return window with a 15% restocking fee, so test it early. For the launch details, see Otto's announcement and IT Brief's coverage.