PlugClaw Now Shipping Worldwide: A Thumb-Sized Private AI Computer That Keeps AI Agents Off Your Devices

Turn any phone, tablet, or computer into an AI workstation — the AI does the work on its own hardware and never touches the data on your main device.

SAN FRANCISCO, Sept. 4, 2026 /PRNewswire/ -- TrustKernel, an AI security and privacy computing company, today announced that PlugClaw has begun shipping worldwide, moving from pre-order to general availability. The company also opened free confidential-model inference on newly deployed servers.

PlugClaw is a complete computer in a USB-C stick: 50 × 19 × 8 mm, 15 grams, with a MediaTek Helio G80 processor, up to 6 GB memory and 128 GB encrypted storage, running PlugOS — a privacy-hardened Android system — plus Ubuntu as the agent runtime. It turns any phone, tablet, or computer into an AI workstation, bringing confidential AI computing to consumer hardware.

Using PlugClaw is simple: plug it into a computer, phone, or tablet, and the AI gets to work.

Once connected, the host device acts as display and input. The AI agent, the applications it calls, login credentials, and the entire execution environment all run on PlugClaw's own hardware and operating system. The agent can open applications, read and reply to messages, organize files, browse the web, and carry out multi-step tasks on its own — but it never needs, and cannot obtain, administrator access to the host device, and has no path to the photos, contacts, documents, or account data stored there.

PlugClaw starts at US$149, a one-time purchase with no mandatory subscription.

Why AI Needs Its Own Hardware to Get Real Work Done

Today's AI agents are already capable enough to take on real work. But to actually let one act, users typically have to hand over device permissions, accounts, and data all at once.

Running an autonomous, self-operating program on your own phone or computer means granting broad access to software whose behavior is not fully predictable.

The industry has already seen a string of public incidents in which AI agents overstepped their permissions, carried out unintended commands, and put user data at risk. The problem was never that agents don't work well — it's that there was no sufficiently secure environment for them to run in.

TrustKernel's answer is to give AI a computer of its own. The agent runs entirely inside PlugClaw's hardware, so its blast radius is confined to the device itself: even if the agent behaves unexpectedly, the impact stays contained within PlugClaw. Unplugging the device ends the session instantly, and the host phone or computer is left untouched.

On the capability side, PlugClaw runs Android natively alongside OpenClaw, an open-source agent framework, extended with built-in GUI-agent and Android-Use capabilities. That lets the AI operate apps within Android the way a person would, completing complex, cross-app tasks.

Thinking Isn't Enough. AI Needs to Act

Most AI assistants only talk. PlugClaw is built to act.

Inside PlugOS, the built-in AI agent doesn't quietly pull data through an API in the background — it behaves like a real operator, opening apps, tapping buttons, typing text, and swiping the screen the same way a person would use a phone. Replying to messages, reading content, organizing files, laying out a slide deck, searching the web: tasks that once required your own hands, PlugClaw completes inside its own operating system, with no developer integration required and no dependence on whether a given app was built to be "AI-friendly."

All the user has to do is say what they need. "Reply to my unread messages in my usual tone," or "put together a competitor analysis deck that matches our brand style" — and PlugClaw opens the right app, reads what's on the screen, and works through it step by step until the task is done, instead of handing back a suggestion and leaving the rest to you.

Powerful AI and Real Privacy Without Compromise

AI users have long faced the same core tension: the more powerful a model is, the more they want to use it — and the more sensitive their data is, the less they dare to. PlugClaw addresses this with cloud-based confidential AI.

On PlugClaw, a user's applications, AI memory, workflows, and sensitive data are all encrypted and run entirely within the device's own independent system — never passing through the host phone, and never stored on a computer's hard drive.

When a task calls for a more powerful frontier model in the cloud, PlugClaw makes that call, but the security boundary holds. Every request is processed inside a hardware-encrypted Trusted Execution Environment (TEE) in the cloud, using remote attestation and memory encryption. Under this setup, neither the cloud provider nor TrustKernel can read the underlying plaintext data, keeping the privacy boundary consistent across the entire task.

PlugClaw gives users a choice between confidential and non-confidential models. Data from preview testing showed that even when both options were available and the non-confidential model was cheaper, users chose the confidential model far more often. Based on that feedback, TrustKernel has deployed dedicated confidential AI servers, offering users free confidential-model inference.

PlugClaw is a one-time hardware purchase with no mandatory subscription. Users can pay as they go with the advanced models pre-installed on the system, or connect their own model service seamlessly through BYOK (bring your own API key).

Thumb-Sized, With an Entire AI Workspace Inside

PlugClaw works with Android phones, iPhones, iPads, Windows, Mac, Linux, and HarmonyOS devices. Plug it into any of them and the AI gets to work. Whichever device a user switches to, their applications, AI memory, custom skills, and workflows travel with PlugClaw, picking up where they left off.

Through PlugClaw's open ecosystem, users get ongoing access to new models, skill packs, and automated workflows.

Who PlugClaw Is For

PlugClaw is built for people who want AI to actually get work done, without exposing the data on their personal phone or computer to do it:

Everyday AI users — People who already use AI to write copy or look things up, but want it to finish the job. PlugClaw works the moment you plug it in, at a fraction of the price of a flagship AI phone.

Privacy-conscious professionals — People who need to use advanced AI on sensitive personal or business information, without exposing their devices, account credentials, or private data.

High-intensity knowledge workers — Consultants, operations, product, marketing, and research professionals who switch between tools all day. PlugClaw can take over repetitive work like organizing meeting notes, tracking to-dos, and completing multi-step tasks across apps. 

Product Specifications

About TrustKernel

TrustKernel is a security computing company focused on secure operating systems, Trusted Execution Environments (TEE), confidential computing, and privacy-preserving technology. Its technology forms the security foundation for a wide range of smartphones, connected vehicles, IoT devices, and enterprise platforms, protecting more than one billion devices worldwide today.

TrustKernel's consumer hardware line began with PlugMate, a thumb-sized private computer for securely protecting personal data and applications. PlugClaw extends that platform into the world of AI agents.

Product page: https://plugos.net/plugclaw

Review units are available to press and creators on request:
TrustKernel
[email protected]

SOURCE TrustKernel