ALIBABA CLOUD × TIDB × ZSTACK Enterprise AI Forum
Enterprise AI · 20 min session
📚 Knowledge
🤖 Agents
🚦 Gateway
🧰 Ops & Governance
OPENTREK
Full-Stack Enterprise AI, Fully Under Your Control
Build, run and govern AI agents — one platform, any infrastructure.
Standalone software · on-prem or hybrid · ZStack-ready
use → / space to navigate
The problem
AI pilots everywhere. Enterprise value still rare.
Most enterprises have bought GPUs or tried assistants — few have production AI the business can rely on.
🧪
Stuck in pilot
DIY RAG hits an accuracy ceiling; POCs never survive to production.
🎇
Shadow AI sprawl
Ungoverned personal assistants — no audit, no quotas, knowledge never shared.
💸
Compute ≠ revenue
GPUs sit idle or burn budget with no token-level visibility or billing.
🔓
Data walking out
Enterprise knowledge leaves the building inside every external prompt.
You don't need more AI tools. You need an AI factory.
The Opentrek idea
One platform. The full AI value chain.
⚙️
GPU TO TOKEN OPERATION
Any GPU becomes a metered, sellable model service — managed, secured, billed per token.
📚
KNOWLEDGE
Scattered documents become accurate, citable answers your people can trust.
🤖
APPLICATION
Loose assistants become a governed digital workforce — hired, trained, managed like employees.
Enter at any point ➜ expand to the full stack — one vendor, compounding value
The platform
Five modules. One control point.
Where it sits · Runs anywhere
The missing middle layer — one software, every infrastructure.
1 · Apps on top
Document writing, coding, data insights, finance, healthcare — the ecosystem your people actually use.
2 · Opentrek in the middle
The enterprise-grade one-stop MaaS platform: knowledge, agent development, runtime, gateway, O&M.
3 · Any infrastructure below
Apsara Stack, public cloud, AI stack, bare metal — and ZStack.
Standalone software. No public-cloud dependency.
🧬 Same software everywhere
🔒 Data & models stay on-prem
🔁 Hybrid cloud, one portal
Better together
ZStack + Opentrek = a private AI factory in one box.
🤖 YOUR AI APPLICATIONS
digital employees · knowledge assistants · AI services
▲▲▲
OPENTREK
models · knowledge · agents · gateway · governance
▲▲▲
ZSTACK
hyper-converged compute · virtualization · simple ops
🏦 Regulated by design
Data, models and agents never leave your premises — sovereignty built in, not bolted on.
🧩 Turnkey stack
ZStack delivers the cloud, Opentrek delivers the AI layer. One stack to buy, deploy and run.
📏 Right-sized
Start on a small HCI cluster; scale to heterogeneous GPUs — NVIDIA and beyond — under one portal.
Use case 1 · GPU to Token Operation
Turn compute into a token business.
📡 Telcos & GPU centers🧠 Model service providers💼 ISVs & IT providers
👨💼
GPU-center
product lead
“Same GPUs as last year — now a branded AI service, billed per token.”
LIVE TENANT METERS · TOKENS / DAY
Gov tenant
1.2M
Edu tenant
640K
Enterprise
310K
0 person-day
White-label portal launch
time to embed your own branded AI portal
sec-level
Token metering
per model · per tenant · per token, with cascading quotas
Any
GPU vendor
NVIDIA and beyond — one token meter across heterogeneous compute
“Owning compute is not the same as monetizing it.”
Use case 2 · Knowledge
Knowledge that answers — accurately.
For finance, government, manufacturing, healthcare — anywhere documents are the business.
What are this quarter's SME lending criteria?
🤖 Max LTV is 70% for manufacturing SMEs — the collateral list was updated in March.
📎 SME_Credit_Policy_v3.pdf · §4.2
12 → 1
Manual steps compressed
one guided flow from upload to trusted answer
0%+
Efficiency gain
vs. manual knowledge processing
0+
Formats parsed faithfully
cross-page tables, formulas, charts — nothing lost
“Anyone can build RAG. Accuracy is the product.”
Use case 3 · Application
A governed digital workforce — not 10,000 loose assistants.
2–4 weeks
PoV on one role
see results on your own workflows before scaling
7×24
Standby workforce
digital employees that never sleep, always audited
0%+
Lower cost with governance
token quotas, shared skills & assets, human-in-the-loop
🤖
Claims agent
drafted 12 overnight reports · 0 policy misses
👩💼
Ops manager
approved 11 · edited 1 · 8 minutes total
✅ human-in-the-loop
Outcomes in the wild
Knowledge that works where accuracy is law. (names anonymized)
💊 Global Top-10 Pharma
PV · ChatBI · KB
Drug-safety literature review · conversational BI · enterprise knowledge base
👩🔬
“Each safety report went from an hour to minutes — with every adverse reaction caught.”
🏦 Note-Issuing Bank · Hong Kong
Credit risk
Credit assessment reports — from multi-source data to analyst final draft
👨
“Data gathering used to eat half the report. Now a draft arrives in an hour — every conclusion cited.”
weeks → days
per new industry
Metrics from production deployments · customer names anonymized
Outcomes in the wild
Every call checked. Every document traced.
🌐 Global Top-Tier Bank
Business QA
Multimodal quality assurance over RM calls, chats and video meetings
👩💼
“We went from checking 1 in 10 calls to every single interaction — in three languages.”
10% → 100%
interactions assessed
1h → 15min
per call reviewed
🚢 International Bank · 160y in Asia
Trade finance
Letter-of-credit examination — senior examiners' logic codified as reusable skills
👨💼
“The AI clears the routine majority; my judgment goes to the cases that matter.”
24×
review speed · 6h → 15min
65%
docs auto pre-screened
Same pattern everywhere: prove on the customer's own data, then scale to the platform.
Why Opentrek
Four reasons business leaders choose it.
🧱
One-stop
Compute → knowledge → application in a single platform. Enter anywhere, expand anytime.
🎯
Accurate
High-fidelity parsing, multi-path recall, built-in evaluation — proven on your documents.
🛡️
Governed
Token quotas, sandbox isolation, full audit, SSO — AI your compliance team approves.
🌍
Anywhere
Standalone on ZStack, bare metal, private or public cloud. Your data stays home.
Not a point tool. Not a sample project. A factory.
Get started
Start small. Prove fast. Scale everywhere.
1
Pick one use case
Knowledge assistant, token business, or a digital employee for one role.
➜
2
PoV on your data
2–4 weeks, on your own documents and workflows — measure the gap.
➜
3
Productionize & expand
Governed rollout on ZStack, then grow across the full value chain.
Let's build your AI factory — on ZStack. 🏭
Thank you · Q&A