1. Home
  2. CA Apps & Cloud services
  3. CA Services
  4. Open-source AI local build service

Open-source AI local build service

Reference:

Condition: New product

03/09/2026

11/04/2026

Brand:

By buying this product you can collect up to 104 loyalty points. Your cart will total 104 points that can be converted into a voucher of $ 520.00.


$ 56,000.00

  • Teams of 1-5 people, 2h/day/user @~30token/s
  • Teams of 6-12 people, 2h/day/user @~30token/s

門市到付 / 過數易 / 貨到付款
門市到付 / 過數易 / 貨到付款

More info

Open-Source AI Local Build Service

Designed for enterprises with confidentiality requirements · Runs entirely on your internal GPU servers · Data does not pass through third parties

Zero Data Leakage

All computation and temporary storage occur on the company's internal GPU servers and are never handled by third parties, meeting strict confidentiality compliance.

Dedicated Local AI Server

The platform can connect to multiple local GPU servers, with linearly scalable computing power.

Internal Exclusive Platform

Preloaded with an Open WebUI / LM Studio style portal, with a customizable dedicated domain.

Up to 120B Class Models

Supports open-source models up to 120B (120 billion) parameters

Up to 128K Context Length

Can process approximately 128K ultra-long documents, code, and reports, meeting deep analysis and long conversation scenarios.

Freely Switch/Connect Multiple Models

One-click switch between open-source LLM/VLM in the management interface, connect multiple model backends simultaneously for output comparison.

* Recommended sweet spot (approx. 80B, 128k context length)
* 80B model averages about 35 tokens/sec, suitable for groups under 10 people.

Dedicated AI Hardware + Software Bundle
1-5 Person Team

Package 1

  • Dedicated AI platform with second-level domain
  • Provision and install Blackwell G10 128GB unified memory
  • Supports running open-source LLM/VLM 80B or below
  • 1-year local hosting (4 vCPU / 16GB / 100GB / no GPU)
Top-level domain must be purchased separately (approx. $300/year)
Supported Model List (Below 80B)
Llama 4 Scout Qwen‑3.5 / Wenn‑2.5 family Mistral Large 2 / Mixtral Command A / Command R+ gpt‑oss‑120b / gpt‑oss‑117B DeepSeek‑V2 / V3 trimmed variants Cohere / Command family variants RedPajama / Mosaic community 65–70B ports & 120B composites Mixtral-style ensembles
6-12 Person Team

Package 2

  • Dedicated AI platform with second-level domain
  • Provision and install 2 x Blackwell G10 128GB unified memory + stacking cable
  • Supports running open-source LLM/VLM 80B or below
  • 1-year local hosting (4 vCPU / 16GB / 100GB / no GPU)
Top-level domain must be purchased separately (approx. $300/year)
Supported Model List (Below 80B)
Llama 4 Scout Qwen‑3.5 / Wenn‑2.5 family Mistral Large 2 / Mixtral Command A / Command R+ gpt‑oss‑120b / gpt‑oss‑117B DeepSeek‑V2 / V3 trimmed variants Cohere / Command family variants RedPajama / Mosaic community 65–70B ports & 120B composites Mixtral-style ensembles

Reviews

No customer reviews for the moment.

Write a review

4 other products in the same category: