PlugClaw Ships Worldwide From $149
PlugClaw ships worldwide from $149, offering a USB-C AI computer that runs agents separately from users’ main devices.

TrustKernel has begun worldwide shipping of PlugClaw, a USB-C computer designed to run AI agents separately from a user’s main phone, tablet, or computer. The 15-gram device measures 50 × 19 × 8 mm, starts at US$149, and includes US$15 in AI credits with no mandatory subscription. PlugClaw uses its own encrypted storage, operating system, applications, credentials, and agent runtime while the connected device provides the display and input. TrustKernel has also opened free confidential-model inference on newly deployed servers.
A Separate Computer For AI
PlugClaw is a complete computer in a USB-C stick, with a MediaTek Helio G80 processor, 4 GB or 6 GB of memory, and 64 GB or 128 GB of hardware-encrypted storage. It runs PlugOS, a privacy-hardened Android system, alongside Ubuntu for the AI agent runtime.
Once connected, the AI agent can open apps, read and reply to messages, organize files, browse the web, and carry out multi-step tasks inside PlugClaw. TrustKernel says the agent cannot obtain administrator access to the host device or access photos, contacts, documents, or account data stored on it.
Unplugging the device ends the session, while the connected phone, tablet, or computer is left untouched. The product is compatible with Android phones, iPhones, iPads, Windows, Mac, Linux, and HarmonyOS devices.
Agent Tasks Run Inside PlugOS
PlugClaw runs Android natively alongside OpenClaw, an open-source agent framework that TrustKernel has extended with GUI-agent and Android-Use capabilities. The setup is intended to let the AI operate Android apps by opening screens, tapping buttons, entering text, and swiping through interfaces.
The product is aimed at tasks such as replying to messages, organizing files, searching the web, and building slide decks without requiring a developer integration or an app built specifically for AI use. Users can move between compatible devices while keeping their applications, AI memory, custom skills, and workflows on PlugClaw.
Confidential Models And BYOK
TrustKernel says PlugClaw encrypts applications, AI memory, workflows, and sensitive data within its independent system. For cloud-based models, the company says requests are processed in a hardware-encrypted Trusted Execution Environment using remote attestation and memory encryption.
Users can select confidential or non-confidential models. TrustKernel said preview testing found that users selected confidential models more often, even where a cheaper non-confidential option was available, prompting the company to deploy dedicated confidential AI servers with free confidential-model inference.
The device supports built-in models including Claude, GPT, Gemini, DeepSeek, Grok, MiniMax, Qwen, Kimi, GLM, and Mimo. Buyers can also pay as they go for advanced models or connect their own model services through BYOK, meaning bring your own API key.
Who PlugClaw Is For
PlugClaw is positioned for everyday AI users who want an agent to complete tasks rather than only provide suggestions. It is also intended for privacy-conscious professionals working with personal or business information, and consultants, operations, product, marketing, and research workers handling repetitive cross-app work.
The one-time purchase model may suit buyers who want to avoid a required subscription while retaining the option to use their own API keys.
A Portable Agent Environment
PlugClaw combines a separate AI runtime with compatibility across phones, tablets, and computers. Users can keep their apps, AI memory, workflows, and custom skills on the USB-C device while choosing between built-in models, pay-as-you-go access, or their own API keys.
From an announcement by TrustKernel.


