TouchVision on-premise LLM appliances let organisations run large language models and AI agents inside their own network. Prompts, documents and outputs never leave the premises, there are no per-token API fees, and the hardware ships pre-configured so teams can start using it on day one.
Three form factors
- YZ-B033 Edge LLM Appliance: a 17 W BM1684X device (32 TOPS INT8) pre-adapted for DeepSeek and Qwen models. It handles private knowledge-base Q&A over PDF, Word, PowerPoint, text and image files, speech-to-text, text-to-speech, text-to-image and code generation, and can also run 32-channel video analytics.
- AI Super Agent: a palm-sized hub for private enterprise models with ASR/TTS, 2D/3D digital humans, meeting minutes, agent assist, natural-language data Q&A and a dashboard designer, with MCP and business API integration.
- YZ-S511 OpenClaw Multi-Agent Server: a workstation-class server with 256 GB DDR5 and 1 TB of AI memory that runs a 229-billion-parameter model and up to 16 AI agents with ready-made business skills.
Where on-premise AI makes sense
On-premise AI is a good fit when data is confidential (contracts, designs, customer records), when the site has limited or no internet access, or when usage is heavy enough that API costs become significant. Typical applications are internal knowledge assistants for manuals and procedures, customer-service and sales agents, document and contract processing, and digital-human receptionists on a self-service kiosk.
Hardware plus integration
Because TouchVision also makes touch displays, kiosks and embedded boards, an LLM appliance can be delivered as part of a complete terminal, for example a voice-enabled information kiosk or a factory HMI that answers questions from equipment manuals. Talk to us through the RFQ form about model selection, languages and integration.
Availability note: some appliances use components and operating systems aimed at the Chinese market. Confirm export configuration, language support and certifications for your country before ordering.
Products in this category
- YZ-B033 On-Premise LLM Edge Appliance (BM1684X, 32 TOPS): A 17 W edge appliance (BM1684X, 32 TOPS INT8) that runs DeepSeek and Qwen models and a private knowledge base fully on-premises, and can also analyse 32 video streams.
- YZ-S511 OpenClaw Multi-Agent AI Server: An on-premise AI agent server that runs a 229B-parameter model and up to 16 concurrent AI agents with pre-built business skills, keeping all data in-house.
- AI Super Agent Palm-Size Enterprise AI Hub: A palm-sized enterprise AI hub for private LLM deployment with speech, digital-human, meeting-minutes, data Q&A and dashboard tools, plus MCP/API integration.
Request a quote for standard or customized versions.
Frequently asked questions
What is an on-premise LLM appliance?
It is a dedicated hardware device that runs a large language model locally inside your network, so prompts and documents are processed without sending data to a public cloud service.
Which models run on TouchVision LLM appliances?
The YZ-B033 is pre-adapted for DeepSeek and Qwen families; the YZ-S511 runs a 229B-parameter MiniMax model. Other open-source models can be evaluated on request.
Do I pay per token?
No. After purchasing the hardware there are no per-token or API subscription fees; operating cost is mainly electricity.
Can an LLM appliance power a digital-human kiosk?
Yes. The AI Super Agent provides ASR/TTS and 2D/3D digital humans that can run on a TouchVision touch display or kiosk.