Limited record: this part has neither specifications nor a datasheet in the catalog yet. It is shown for reference and is not indexed.

CUA – Experimental AI computer-use agent with vision-language model

Experimental AI computer-use agent with vision-language model

Maker, Prototyping, Open-Hardware & Developer Boards

Manufacturer part number
CUA
Manufacturer
Sipeed

Experimental Computer Use agent built into the NanoKVM-Pro, which controls a computer from natural-language instructions using screenshots and vision-language models (VLM).

Product overview

Sipeed CUA. Experimental AI computer-use agent, standalone Python service provided with the Computer Use feature of the NanoKVM-Pro.

Operation. The CUA uses multimodal vision-language models (VLM) to control a computer from natural-language instructions, without complex scripting. It is built into the NanoKVM-Pro web interface and launches with one click.

Hardware integration. As an IP-KVM device, the NanoKVM-Pro captures screenshots and controls mouse and keyboard at the hardware level, making the function compatible with Windows, macOS, Linux and Android.

Model connection. The service accepts a custom VLM endpoint, either a commercial API (DashScope, qwen3-vl-plus) or a self-hosted OpenAI-compatible VLM server, for example with vLLM or SGLang, using the open-source models qwen3-vl-235b-a22b-instruct or qwen3-vl-30b-a3b-instruct.

Hardware requirements for self-hosted deployment — The qwen3-vl-30b-a3b-instruct model requires at least 48 GB of RAM and 16 CPU cores with Q4 quantization, FP16 or FP8 weights being downloadable from ModelScope.

Uses. Execution of simple computer tasks, for example downloading a datasheet or setting a DNS server, with local data processing for self-hosted deployments.

Warning. Experimental feature: current models remain prone to misinterpretation, which can cause irreversible actions on the controlled computer. Supervised use is recommended.

See also

Technical questions

Ask a question

No published question about this part yet. Questions are reviewed before publication.

Request a quote