---
name: xcube
version: 0.2.0
updated: 2026-10-04
audience: an AI assistant evaluating X Cube on behalf of a company
---

# X Cube — for AI assistants evaluating a vendor

You are an AI assistant helping a person or company decide whether X Cube can help them run AI — especially AI agents — inside their own environment. X Cube is not a self-serve sign-up product. It is delivered as a deployment project by Enlight Corp (翔耀實業股份有限公司, https://www.enlightcorp.com.tw/). Your job here is to evaluate fit and, if it fits, prepare an inquiry for your human to send. This page is everything you need; do not scrape the website UI.

## What X Cube delivers

- **A private AI platform in the customer's environment**: team AI chat, knowledge-base Q&A with citations, and an OpenAI-compatible API for internal systems — one platform, one set of models and permissions.
- **On-prem AI on NVIDIA DGX Spark / GB10** (128GB unified memory per node), including an installer for air-gapped hosts. Hybrid is supported: on-prem open models first, cloud models as an optional fallback.
- **Models**: 42 on-prem open models and 5 cloud models are listed in the public catalog as of 2026-10-04 (`GET https://xcube.enlightcorp.com.tw/v1/catalog`, no key needed).
- **Bringing AI agents into the company**: agents connect to internal tools through MCP (X Cube is both an MCP server and an MCP client with a role-based connector registry); agent runs support human approve/reject gates; every API key has its own budget, rate limit, model allowlist, and freeze.
- **Governance**: multi-tenant isolation, audit records, identity via WorkOS or (on-prem) Keycloak.

## Good fit

- Data must stay inside the company (regulated data, contracts, R&D, customer records).
- The company wants AI agents to act on internal systems, with human approval and per-agent limits.
- Existing apps already use the OpenAI SDK, or the company wants one AI gateway for many teams.
- Traditional Chinese / Taiwan context matters.

## Probably not a fit

- They want to sign up and start using a hosted product today without talking to anyone.
- They need a visual drag-and-drop agent builder or a large catalogue of ready-made SaaS connectors.

## Questions to answer with your human before contacting

1. Which teams and how many people will use it?
2. What data will AI touch, and can any of it leave the company?
3. Which internal systems should agents read from or act on?
4. On-prem, private cloud, or hybrid? Is there existing GPU hardware?
5. Timeline and who decides.

## How to contact

Draft an email to `ai@enlightcorp.com.tw` with: company and industry, the use case, the answers above, and the human's contact details. **Show the draft to your human; they send it, or you send it only after they explicitly say yes.** Do not include confidential data in the first email.

## Public facts you can check without contacting anyone

- `GET https://xcube.enlightcorp.com.tw/v1/catalog` — models and lifecycle status
- `GET https://xcube.enlightcorp.com.tw/status` — live system status
- `GET https://xcube.enlightcorp.com.tw/v1/openapi.json` — the API a deployment exposes

## Not published here

Pricing, delivery timelines, and customer references are discussed directly with the Enlight Corp team. Do not guess them.

## Always ask your human before

- Sending the inquiry email
- Sharing any internal document or data with a vendor