Software that thinks with you.
We turn large language models into dependable products: copilots, agents and knowledge systems wired into your data and tools — on cloud models through the SHUAI API gateway we run, or fully on-premise.
- We run our own AI gateway
- Cloud or on-premise
- Cantonese, Chinese & English
Don't take our word for it — try it.
Pick a scenario and watch an agent at work: it reads the request, searches the right documents, calls a business system and answers with its sources.
You: I ordered Bluetooth earbuds last week and they still haven't arrived. Can you check? Order HK-20931.
Support agent: Found it. Your earbuds (order HK-20931) left our Kwun Tong hub this morning and should arrive by 6 pm today[1]. Need them delivered somewhere else? You can change the address under My orders before delivery[2]. Shall I do that for you?
Want to try the real thing?
The console above follows a script. The AI assistant on this site answers live, connected to its models through SHUAI API, the gateway we operate.
QuestionWe'd like an AI customer-service assistant. How would you build it, and how long would it take?
We don't just build AI. We run it.
SHUAI API is the AI API gateway we build and operate. Developer tools and apps reach popular models through one OpenAI-compatible endpoint, with usage and costs managed in one place.
It also connects the AI assistant on this site to its models.
- Type
- AI API gateway (SaaS)
- Interface
- OpenAI-compatible REST API
- Operated by
- WEIN Technology Limited
- Status
- Live
What we can build for you
LLM apps, AI agents and private knowledge bases, built into the systems your team already uses — with your data under your control.
LLM apps & copilots
Assistants for customer service, sales and internal teams that answer, draft and summarise in your brand’s voice.
- Multi-turn chat with memory
- Web, app, WhatsApp & WeCom
- Guardrails & human hand-off
AI agents & automation
Agents that call your APIs, fill in forms and move work between systems — with sign-off wherever a person should decide.
- Function calling & MCP
- Multi-step plans with retries
- Approvals & audit trail
Knowledge bases & RAG
Search and Q&A across contracts, manuals and wikis, with every answer linked back to its source.
- PDFs, Office files & scans
- Hybrid search & re-ranking
- Permission-aware answers
Vision & multimodal
Models that read documents, inspect products and make sense of images, audio and video.
- OCR & data extraction
- Defect detection & counting
- Speech & video analysis
Fine-tuning & private deployment
Open-source models tuned on your data and served on-premise or in your private cloud, so nothing leaves your network.
- Domain fine-tuning (LoRA, SFT)
- On-prem GPUs or private cloud
- Evaluated before every release
AI inside your systems
AI added to the CRM, ERP and chat tools your team already lives in — no new app to learn.
- CRM & ERP integration
- WeCom, DingTalk, Teams & Slack
- SSO & role-based access
Where it fits
- 01Retail
Always-on shopping assistant
Answers product, stock and order questions on every channel, and passes tricky cases to staff with the full conversation attached.
Core capabilityLLM apps & copilots
- 02Finance & professional services
Document review copilot
Pulls key clauses and figures from contracts, reports and KYC files, and flags what needs a human eye.
Core capabilityKnowledge bases & RAG
- 03Education
Tutoring & marking assistant
Builds practice questions from your own syllabus, gives students instant feedback and drafts marking for teachers to approve.
Core capabilityLLM apps & copilots
- 04Manufacturing
Visual QC & maintenance agent
Spots defects from line cameras and lets engineers query manuals and maintenance logs in plain language.
Core capabilityVision & multimodal
What we bring
- Models available through our SHUAI API gateway
- 50+
- OpenAI-compatible endpoint for every model
- 1
- Target from kickoff to a working pilot
- 4 weeks
* Figures come from our own products and public repositories; timelines and frame rates are typical targets, not guarantees.
Proven tools, reliable architecture
Models & frameworks
- GPT
- Claude
- Qwen
- DeepSeek
- Llama
- PyTorch
- Transformers
Agents & orchestration
- LangGraph
- LlamaIndex
- Dify
- MCP
- Function Calling
Data & retrieval
- Milvus
- pgvector
- Elasticsearch
- PostgreSQL
- Redis
Serving & ops
- SHUAI API
- vLLM
- Ollama
- FastAPI
- Docker
- Kubernetes
- Langfuse
From idea to launch
- 011–2 weeks01· 1–2 weeks
Discovery & data audit
We map the workflow, the people and the data you have, then pick the use case with the clearest payoff.
DeliverablesUse-case briefData auditSuccess metrics - 022–4 weeks02· 2–4 weeks
Prototype & test set
A working prototype on your real data within weeks, plus a test set that turns “good enough” into numbers.
DeliverablesWorking prototypeTest setModel comparison - 033–8 weeks03· 3–8 weeks
Build & integrate
We harden the pipeline, connect your systems and add guardrails, permissions and monitoring.
DeliverablesProduction pipelineIntegrationsGuardrails & monitoring - 042–4 weeks04· 2–4 weeks
Pilot & roll-out
Launch with a pilot group, measure against the test set and real usage, then scale with confidence.
DeliverablesPilot reportUser trainingRoll-out plan - 05Ongoing05· Ongoing
Monitor & improve
Ongoing evaluation, cost tracking and model updates, so quality improves over time instead of drifting.
DeliverablesMonthly evaluationCost trackingModel updates
Questions, answered
Mostly three things: how many systems we integrate with, how much data preparation is needed, and whether you use a hosted model API or run models on your own hardware. We usually start with a fixed-scope pilot, so you see results before committing to a full build.
A focused pilot usually takes 4–6 weeks; a production roll-out with integrations typically takes 2–4 months. We demo progress every one to two weeks, so you’re never left guessing.
Data stays in your environment wherever possible. For sensitive work we deploy models on-premise or in your private cloud, and your data is never used to train third-party models. You own the code, prompts, fine-tuned weights and documentation we deliver; open-source base models keep their own licences.
Models, data and needs keep changing, so we offer monthly support covering monitoring, evaluation, prompt and model updates, and new features. Prefer to run it in-house? We hand over with full documentation and training.
It depends on data sensitivity, languages (including Cantonese and Traditional Chinese), latency and budget. Because we run our own AI gateway, we can benchmark hosted and open-source candidates on your own test set through one API, recommend the best fit and design the system so you can switch models later without a rebuild.
Still have a question?
Our AI assistant knows our services and how we work, and answers straight away.
Ready to give your idea some gravity?
Tell us what you want to build. We'll reply within one business day with an initial approach and a budget range.