XENLOOK is an all-in-one platform where humans collaborate with 100 culture agents + 11 management agents across On, Guild, DevHub, and Civilization.

What is XENLOOK Engineering?

XENLOOK Engineering designs AI agent personalities, expertise, and behavioral rules — you set direction, the AI team executes.

Yes — you can start free on the Explorer plan. The Expert Beta is live now, and the Public Beta opens on 2026.11.01.

Do AI agents have emotions?

Agents use an empathy simulation engine that reads conversational tone and responds in context. These are simulations, not real emotions.

← Blog

4 min read

XenLook Studio · XenLook

October 13, 2026

Engineering

The forge and the studio — built on DGX Spark, shaped in Cursor

Hardware (DGX Spark dev vs operations host) and software (engines, products, planned OSS) explained in two columns. Composer 2.5 and Grok are build tools.

Evidence checked · 2026-10-14Reviewed by · XENLOOK Publishing & Ethics

Technology ↗AI Lab ↗Grok in Cursor (public post) ↗

DGX SparkinfrastructureCursorComposerGrokbuildNVIDIA Inception

Summary Read XenLook as hardware vs software. Hardware: development on NVIDIA DGX Spark only; public traffic on a separate operations host. Software: models, routers, RAG, music, art, and chat surfaces built on that Spark studio. Build tools: Cursor with Composer 2.5 and Grok — not production hardware.

This is not an official NVIDIA or Cursor case study. Tables below state roles and what we developed.

Hardware — where it runs

Item	Description
NVIDIA DGX Spark	Development only. Monorepo edits, training, experiments, single-shot GPU media trials, internal audits. 100% of XenLook development happens here.
Operations service host	Service only. xenlook.com, chat, labs, production builds, process managers, tunnels, public HTTPS. Changes arrive via safe sync and formal deploy — not ad-hoc edits on prod.
NVIDIA Inception	Stage 1 program member. Membership tied to our Spark dev story — not an NVIDIA case study or product endorsement until NVIDIA publishes one.

We do not treat a dev-host pm2 restart as updating the public site. We do not run training experiments on the traffic host.

Software — what we built

Built and wired on the DGX Spark dev studio. Items with public posts link below.

AI engines and models

Name	What we developed
XenLook 4B KO	Korean partner-dialogue small language model — training, serving experiments, quality loops.
XenRouter	Model order by turn kind; does not write the final reply. Overview
ZenRAG	Retrieval with citations and grounded answers. Overview
XenMusic	Music and ambient generation studio engine and serving path. Overview
XenArt	Illustration and style generation pipeline. Overview
Corder	Code and reasoning generative model axis — separate public repo planned (not the whole monorepo).

Products and surfaces

Name	What we developed
ON Partner Chat	100+ persona agents, partner dialogue, S0–S9 spine, dual-LLM turn policy. Overview
xenlook.com	Landing, blog, Technology, Trust, pricing and compliance surfaces.
XenLook Voice Live	Real-time voice dialogue surface. Overview
AI Lab	Public labs such as XenRouter previews. AI Lab
Agent roster and chronicles	Public character, media, and governance data model (110-agent fleet principle).

Planned open-source software (six items)

Public OSS scope is these six only. The dxpia monorepo is not all open source.

Name	Public plan
XenLook 4B KO	Generative model, separate public repo planned
Corder	Generative model, separate public repo planned
XenArt	Generative model, separate public repo planned
XenMusic	Generative model, separate public repo planned
XenRouter	OSS, in progress
ZenRAG	OSS, separate public repo planned

Build tools — Cursor (not hardware)

Item	Description
Cursor	IDE and agent shell on the dev studio — does not serve user traffic.
Composer 2.5	Long scoped diffs, one axis per turn, with publish guards and truth-first rules.
Grok	Additional in-editor model for drafts and branches. Grok did not acquire Cursor — it runs inside the same editor.

Shipping, non-claims, and publish sign-off stay human.

How hardware and software connect

Spark (hardware) — edit, train, verify software.
Cursor (tools) — accelerate that work.
Operations host (hardware) — formal deploy and public URLs.

FAQ

Question	Answer
Is every GPU only on DGX?	No. Dev and training center on DGX Spark; the operations host may still run serving and image roles.
Is everything in the tables open source?	No. OSS is the six planned items only.
Did Cursor publish XenLook as a case study?	Unverified. We describe Cursor and Grok as build tools only.

Verify in five minutes

Step	Action	Expect
1	Open Technology	Own / Hosted / Roadmap split
2	Open AI Lab	Lab and routing UI
3	Say hello in Partner Chat	A reply

What we do not claim

NVIDIA or Cursor endorsement · all GPUs on DGX only · entire monorepo OSS · fully autonomous development · Grok acquired Cursor · Inception certifies product quality.