300+
users onboarded to Nexus Composer: 150 in September 2026 and 150 in October 2026.
Build infrastructure – Secure data – Own your AI
Vietnam’s AI Infrastructure & Platform Company
OneNexus is Vietnam’s AI infrastructure and platform company, developing an end-to-end AI value chain for enterprises with three core focus areas: AI Cloud, Data Platform, and AI Application.
AI Application · Available in pilot
AI that doesn’t just assist — it works alongside you.
Coworking agent for enterprises
Data Platform · On the roadmap
Your data, centralised and under control
AI Cloud · Inference in pilot
Compute and models
01
Flagship product
Nexus Composer
OneNexus’s flagship product, and a true coworking agent. From a request described in plain everyday language, Nexus Composer can plan, get to work, and hand over a finished product — instead of just giving an answer.
Below are real tasks completed during the pilot, running on GLM-5.3-Vision (GLM-5.3-Flash MXFP4). See Nexus Composer →
Request: “build an Excel file to manage personal finances”
2:30 4 sheets, with charts and a budget table.
Request: “make an intro clip for Nexus Composer”
5:08 A 30-second 1080p clip: script, visuals, video editing.
02
Engineering
Models optimised for AMD, recognised by the community
Nexus Composer runs on an AI stack whose technology OneNexus owns end to end — from quantization and optimisation to real-world deployment. Together with RadixArk and AMD, we deliver faster, lighter processing on AMD Instinct MI355X GPUs. Try it now — completely free, with unlimited tokens during the trial period.
The MXFP4 checkpoint for GLM-5.3-Flash on AMD MI355X
The SGLang Cookbook uses OneNexus/GLM-5.3-Flash-MXFP4 as the MXFP4 checkpoint in its GLM-5.3-Flash deployment recipe for AMD MI355X.
MXFP4 builds of GLM-5.3 and GLM-5.3-Flash for AMD GPUs
OneNexus publishes MXFP4 quantizations of GLM-5.3 and GLM-5.3-Flash on Hugging Face. The GLM-5.3-Flash release ships with its measurement protocols, so you can reproduce the results.
−40%
checkpoint size: 305.78 GiB → 182.12 GiB
+12.4%
output throughput, geometric mean over c1–c48
+18.2%
P90 output speed per user
+24%
on-device KV cache: 13.97M → 17.32M tokens
OneNexus GLM-5.3-Flash MXFP4 versus the official GLM-5.3-Flash FP8 checkpoint on the same four MI355X GPUs (geometric means, c1–c48). Against AMD’s own MXFP4 checkpoint, serving speed is at parity. 11,810 requests across eight AgentX load levels, with zero request errors. See how we measured and reproduce the results
03
Vision
Own your AI, from the foundation up
AI has become a core capability, reshaping how businesses operate, make decisions and create breakthrough value. As AI systems depend deeply on infrastructure, data and operational tooling, a business cannot stop at “using AI” — it needs full control, and the autonomy to build and grow its own AI capability.
AI Cloud
Compute platform Inference in pilotCompute infrastructure and models optimised for training, inference and every AI workload — scaling easily and flexibly with the size of your business.
Data Platform
Data foundation On the roadmapCentralised data connection and governance on a single platform, keeping data secure and giving AI a solid foundation.
AI Application
Operational applications In pilot with Nexus ComposerBring AI from the platform into your processes, operations, and everyday work, creating direct, measurable value for the business.
Working alongside businesses in Vietnam and the region, OneNexus is committed to being a long-term partner in building AI capability that is self-directed, easy to scale and optimised over time.
Own your AI capability. Build lasting value.
04
Deploy
Deploy AI flexibly, your way
Use it now · Integrate deeply · Customise your own
Available in pilot
01 · WORK A smart work assistant
Nexus ComposerHand off work in plain language. AI plans, executes and completes the work end to end.
Available in pilot
02 · BUILD AI infrastructure for your products
Nexus InferenceOne unified API connects you directly to foundation models, speeding up enterprise application development. Or easily deploy your own models (self-hosted) on a dedicated endpoint.
Private preview
03 · CUSTOMIZE Specialised AI models
Nexus TrainingTrain and fine-tune models on your company’s proprietary data, ensuring the highest accuracy for every line of business.
05
FAQ
Frequently asked questions
Does the pilot cost anything?
No. 100% free for the whole trial. Trial period: Expected to run 4–6 weeks (exact dates agreed at kickoff). Runs on shared infrastructure with a fixed quota. Issues that come up are picked up and handled during support hours.
Are my prompts and data stored?
100% of your prompts and outputs are never stored on our systems, and never used to train or fine-tune any OneNexus model. Every workspace, API key, quota, piece of data and usage record is partitioned per organisation, fully preventing cross-tenant access.
Who can join the pilot?
Teams within One Mount Group and Techcombank have priority. External teams can register interest and hear as soon as access widens.
How do I install and use Nexus Composer?
Install steps, system requirements, and downloads are on the Nexus Composer page.
Who do I contact if something goes wrong?
Contact your account’s onboarding specialist directly — the person who handed over your workspace and issued your API keys. Support hours: 09:00 – 18:00 (GMT+7), Monday to Friday.
Request pilot access
Share your challenge with us
We’ll work with you to find the right starting point, and get back to you with a deployment plan within 1 business day.
See who qualifies, what it costs, and how your data is handled on Pilot & Security.
Already in the pilot? Get Nexus Composer
💡 For customers outside One Mount Group & Techcombank: Please choose “External interest” — we’ll notify you as a priority as soon as we open the pilot to more teams.