n8n · Dify · custom code

AI workflow automation with n8n and Dify, built in Hong Kong

We design, build and hand over AI-powered workflows for Hong Kong organisations: enquiry assistants in Cantonese and English, documents that turn into structured data, notifications and approvals that run themselves, knowledge assistants that cite their sources. Built on n8n, Dify or plain code, deployed on your own server or cloud; the source and documentation are yours at the end.

Tell us on WhatsApp See how it works

n8nDifyPython custom nodesDocker / KubernetesSelf-hosted or cloudCantonese and English

A Hong Kong studio desk at dusk: a monitor showing a workflow of connected nodes, from a chat message through a document and a spreadsheet to an approval tick, with a phone lit by a chat bubble and the harbour skyline behind

What we automate

The workflows Hong Kong offices, NGOs, schools and practices ask for most. Each one connects to the systems you already run rather than replacing them.

Enquiry assistant with staff handover

An AI assistant on your WhatsApp number or website that answers service questions in Cantonese and English from your own material, and goes quiet the moment a colleague replies in the same chat.

Enrolment, payment and reminder notifications

Confirmations, payment notices, waitlist updates and appointment reminders sent automatically from your enrolment or booking records, on WhatsApp or email, at the right time.

Documents in, structured data out

Invoices, application forms, referral letters and scanned PDFs read by AI and written into your spreadsheet, database or accounting system as fields, not as a chat window someone copies from.

Knowledge assistant with citations

Staff ask questions about your SOPs, policies, product sheets or case guidelines and get answers with the source paragraph attached, so every answer can be checked.

Approval and case workflows

Requests routed to the right person, drafted replies waiting for a human yes, status pushed back into your case-management or HR system. AI drafts; people approve.

Scheduled reports and system sync

Nightly exports, monthly summaries in plain language, and two systems that finally agree with each other, without anyone re-keying data on a Friday afternoon.

n8n, Dify, or code: we pick the right one for each job

All three are tools we build with every week. The choice depends on who will maintain the workflow after handover and how much of it is AI conversation versus plumbing.

n8n

Triggers, schedules and cross-system plumbing

Open-source workflow automation that your own staff can read and adjust. We build the nodes, error handling and retries, add Python or JavaScript custom nodes where a connector is missing, and run it self-hosted on Docker or Kubernetes or on n8n cloud.

Dify

Knowledge bases, prompts and chat apps

Open-source LLM application platform for assistants that answer from documents. We set up the knowledge base, retrieval and prompts, connect OpenAI, Anthropic, Azure OpenAI or a locally hosted model, and expose the result to WhatsApp, a website widget or n8n.

Custom code

When the platform would fight you

TypeScript or Python services for workflows with heavy logic, high volume or unusual systems: SQL and ODBC databases, on-premise vendor software, undocumented APIs. Same handover: source, tests, deployment notes.

Deployment is your call: your cloud account, a server on your premises with data staying in Hong Kong, or hosting by us. Whichever platform we use, the workflow ships with production guardrails: scoped credentials, approval steps for anything that writes, budget caps on model usage, and logs you can audit. For MCP servers and deeper agent integration see AI Agents & MCP.

Who we build this for

Organisations that need automation to run reliably without an in-house developer, often with funding rules to satisfy. Examples of the kind of workflow each one asks for:

Social service organisations and NGOs

A Cantonese enquiry assistant for centre services and programme schedules, with staff taking over on WhatsApp when needed
Programme enrolment confirmations, reminders and waitlist notices driven from your existing enrolment or membership records
Case-note summaries and standard letters drafted from your own templates, with a knowledge assistant over internal guidelines

+ Funding this

The SWD Dedicated Fund for System Enhancement - IT Projects (Appendix A2) and other Lump Sum Grant procurement. We prepare the technical description and quotation your application needs.

Schools and education centres

A parent enquiry assistant for admissions, fees and school-bus questions in Cantonese and English, day and night
Circulars and reply slips turned into structured records with automatic reminders to parents who have not responded
A staff knowledge assistant over school policies and procedures, answering with the source paragraph attached

+ Funding this

Education Bureau IT-related grants and the Quality Education Fund where the project qualifies. We write the technical section of the proposal with you.

Clinics and professional practices

Appointment confirmations and reminders on WhatsApp from your booking system, with rescheduling handled in the same chat
Intake forms, referral letters and reports read into structured patient or client records
A knowledge assistant over practice manuals, fee schedules and regulatory guidance that cites what it quotes

+ Funding this

The Technology Voucher Programme (TVP) covers up to three quarters of an eligible project. We supply the project description and quotation in the format the application requires.

Start with a pilot

One fixed price for one real workflow, so you can see it working in your own systems before committing to anything larger.

AI Workflow Pilot

One workflow, one system connected, live in about two weeks.

HK$18,800fixed, one-off

Timeline: About 2 weeks from deposit and access

A short scoping call, then the pilot scope and price confirmed in writing before any work starts
One workflow built on n8n, Dify or in code, connected to one of your systems (form, spreadsheet, database, CRM, WhatsApp or email)
Cantonese and English where the workflow talks to people; human handover where it matters
Deployed on your cloud account or server, or hosted by us for the first three months
Source, credentials handover and a one-page runbook, plus a walkthrough for your team

Ask about the pilot on WhatsApp

50% deposit to start, balance at go-live. Model and messaging usage (OpenAI, Anthropic, Meta) is billed to you at cost with no markup.

What comes after the pilot

Knowledge assistant

A full RAG assistant over your document set, in Cantonese and English with citations, on WhatsApp, your website or an internal tool.

System integration and custom MCP

Workflows across several systems, agents that act inside your CRM, POS or database under approval, and connectors for Claude, ChatGPT or Cursor.

Monthly care

Monitoring, model and platform updates, small changes to prompts and flows, and a person to call when something looks wrong.

Everything beyond the pilot is scoped and quoted in writing after a short call. Most funded projects we take on sit in the tens to low hundreds of thousands of Hong Kong dollars, depending on the systems involved and the volume.

Ranges are not a rate card; you get a fixed figure before any build starts.

Not a pilot-shaped problem, or already holding an RFP? Tell us what you need and we quote it directly.

Prefer email? info@virtualhi.hk

How an engagement runs

1. Workflow audit

A short call or WhatsApp exchange to map one painful workflow: where the data lives, who needs the result, what the AI may touch. You receive a written scope and a fixed price within days.

2. Working workflow in days

A real n8n or Dify workflow running against your actual systems, not a slide. You try it with real cases and we adjust prompts, routing and edge cases together.

3. Production and handover

Credentials scoped, approvals and logging in place, deployed where you chose. Source, runbook and a walkthrough handed over, with monthly care if you want it.

Questions procurement and IT usually ask

The same answers we give on compliance tables and quotation forms, written out plainly.

Can it run on our own server, with data staying in Hong Kong?+

Yes. n8n and Dify are both open source and run self-hosted on Docker or Kubernetes, on a server on your premises or in a Hong Kong cloud region. The AI model can be a hosted API or a locally run model, depending on the sensitivity of the data. We also host on your behalf if you prefer.

Can you connect to our existing case-management, enrolment or accounting system?+

Usually, yes. If it has an API we use it; if it only has a database, exports or a web interface, we work with those. SQL and ODBC databases, on-premise vendor software and undocumented APIs are a large part of what we are asked to connect.

Do we own the source code and documentation?+

Yes. You receive the workflow definitions, custom code, deployment notes and a runbook. Nothing is locked to a platform you rent from us, and any competent developer can take it on afterwards.

What happens when the AI gets an answer wrong?+

We design for it. Knowledge assistants answer with citations so staff can verify. Anything that writes to a system or sends a message to the public goes through an approval step or a human handover. Model usage has budget caps and every action is logged.

How do you handle personal data under the PDPO?+

Data stays in the systems and region you choose; we minimise what is sent to any model, avoid retaining prompts where the provider allows it, and document the data flow so you can include it in your own privacy impact assessment. We sign your confidentiality and data-processing terms.

Do you provide training and a proper handover?+

Yes. Every build ends with a walkthrough for the people who will run it, a written runbook, and a period where we are on hand for questions. Documentation is in English, with Traditional Chinese on request.

What does support cost after go-live?+

Nothing is compulsory. You can run the workflow yourself, call us for ad-hoc changes, or take monthly care covering monitoring, updates and small changes. The monthly figure is quoted with the project so there are no surprises.

We already have n8n or Dify installed. Can you work inside it?+

Yes. We are happy to build inside your existing installation, review workflows your team has started, and harden them for production. The pilot works the same way on your instance as on a fresh one.

Get a quote

Email WhatsApp