Agentic CS
Live · invite-onlyAI customer service that answers only from a business's own documents, in English, Chinese, and Bahasa Malaysia. Live and invite-only, with design partners under signed contracts.
DuoCode AI Customer-Service Operating System
DuoCode builds Agentic CS, an assistant for Malaysian SMEs that answers customers only from the business's own documents, in English, Chinese or Bahasa Malaysia. If an answer isn't there, the conversation goes to a person — it is not guessed. It runs at agenticcs.net.
The company
Our product
Agentic CS is the product this page describes. Below it, the company's other products — eight more we run, all live.
The company's other products
Tradrill
LiveA training platform for reading technology stocks: transmission-chain research notes and a paper-trading simulator. Free to use; it does not place trades or give stock picks.
HipiMind
LiveA voice-first AI travel assistant: plan a trip by talking, then listen to an audio guide for the place. On the web and on Google Play.
FluentDojo
LiveA spoken-English coach for Chinese speakers, used in the browser: talk with an AI coach, then get a review of the conversation.
签屿 VisaIsle
LiveMock interviews for the US B1/B2 visa. It reads an applicant's DS-160 and questions the inconsistencies, in Chinese and in English.
FoldRender
Live on AndroidAn Android document reader and git client for the Galaxy Z Fold, on Google Play, plus a web service for publishing a page and sharing the link. A separate iPhone/iPad app for reading published pages is in public beta on TestFlight.
APRide
LiveA ride-matching board for students and staff of Asia Pacific University, Kuala Lumpur. On the web and on Google Play.
Covate
LiveA free, open-source tool that quizzes a developer on their own code changes, so building with AI still teaches something.
polyagent
LiveA social deduction game played at a table of AI characters. Free, and no sign-up.
Agentic CS · The service loop
Answering is the easy part. Resolving is a loop.
An AI can answer a customer — but the service job is often still unfinished. A useful answer needs the right customer context, the correct approved policy, a safe next action or a human handoff, and a record of what happened.
The gap isn't conversation. It's completion.
So Agentic CS connects the conversation to the work that resolves it — one governed path from request to resolution.
Agentic CSGoverned loop
Ask
Ground
Decide
Act / hand off
Verify
Record
01
Ask
A customer asks — by text, a screenshot, or voice.
02
Ground
Pull the right customer context and answer only from approved, released knowledge.
03
Decide
Apply the approved policy for the case — not a guess.
04
Act / hand off
Take an authorised action, or pass risky work to a human with full context.
05
Verify
Check the evidence — no false promise, and no money moves without approval.
06
Record
Log the outcome to an audit trail that improves the next answer.
The recorded outcome becomes context for the next question — so the loop closes and every answer gets sharper.
See the loop running
The same loop, in the real product
Asked whether refunds are automatic, it searches approved sources and explains the policy with citations — never inventing a limit or promising money moved. The next step is a controlled review, not a false promise.
The same conversation opens to a human — message, answer, sources, and recorded activity. The operator takes over with full context, so the customer never repeats themselves.
The business reviews and releases which documents may answer customers — the AI is never handed a pile of files to improvise. It controls what the AI is allowed to use.
The difference
Most tools optimise for a faster reply. We optimise for a defensible resolution — approved evidence and citations, policy-bound next steps, and human handoff with an audit trail.
We start with Malaysian D2C and SME merchants — typically on Shopee, Shopify, or WooCommerce — where a wrong promise about an order, payment, delivery, or refund is expensive.
Better service should never require a business to surrender control.
Agentic CS is DuoCode's own product, built and operated by us, and running in production at agenticcs.net.
Grounded · policy-bound · improved from audited outcomes — built with Codex + GPT-5.6
One system behind the loop
Six modules, one runtime truth
The loop isn't a demo stitched from parts. Every module writes back to the same product state, audit trail, source governance, and approval workflow — so the whole loop stays consistent no matter which part is doing the work.
CustomerAgentTools
Customer-Agent Runtime
A multilingual support agent that streams answers from approved sources, customer context, orders, payments, inventory, and tenant-scoped tools instead of loose prompt memory.
TraceRiskReview
Gemini Audit Layer
A shadow reviewer over the Convex audit bundle. It comments on safety, evidence, and handoff quality without answering customers or mutating product state.
BriefRunArtifact
Design Studio Debug View
A live product lane for testing website prototype workflows, source-grounded briefs, worker runs, artifact state, and support questions before they become customer-facing features.
SourceReleaseCitation
Evidence & Source Governance
OpenAI File Search provides retrieval, while Convex owns source authorization, releases, provider file ids, citations, deletion jobs, and replayable evidence.
QueueContextApprove
Human Approval Cockpit
Refunds, data access, account changes, and unsupported policy exceptions route to an operator queue with context, trace, source evidence, and one-click decisions.
OrderPaymentDelivery
SME Integrations
Payments, orders, CRM, POS, inventory, delivery, websites, and internal systems connect through audited custom actions instead of hidden automation shortcuts.
Provider-backed evidence
Citations come from real OpenAI File Search payloads and are persisted back to Convex for replay, deletion, and audit.
Shadow review, not auto-control
Gemini reviews the audit bundle and flags risk. It never approves refunds, changes tickets, or replaces operator judgement.
Shipped SME workflow references that inform the product
Workflow Evidence Library
These shipped references are not the main product. They are the operating evidence behind DuoCode's approved business knowledge, support workflows, Design Studio testing, and SME integrations. Use them to see the industries and workflows the product is designed to support.
Built around product truth, not chatbot guesses
A Malaysia-based software company turning SME support, approvals, and workflow design into one inspectable AI product
DUOCODE TECHNOLOGY is a Malaysian software company registered with SSM Malaysia (Reg. No. 202603062356) and headquartered in Klang, Selangor. It builds and runs Agentic CS (agenticcs.net) and the eight other products listed on this page — from a spoken-English coach to a mock visa-interview practice tool.
SSM Registered
Reg. No. 202603062356 (LA0087244-A) · Klang, Selangor
Shared Runtime Truth
Conversations, tickets, tools, citations, run events, and reviews in one trace
Audit-First AI
RAG, tool validation, Gemini review, approval queues, and grounded answers
Local SME Fit
English, Chinese, Bahasa Malaysia · FPX, wallets, marketplace workflows
DUOCODE TECHNOLOGY · sole proprietorship registered in Malaysia on 4 March 2026, no. 202603062356 (LA0087244-A) · No 4 Jalan Mohd Tahir 1, 41200 Klang, Selangor.
Team
Key members and their industry experience.
Liu Wei
Co-Founder & CEO
Computer Science (Hons) at Asia Pacific University Malaysia, on the APU–De Montfort University dual-award programme. Agent and applied-AI engineering across three China-based AI companies: algorithm intern at iMean (Hangzhou), embodied-agent algorithm intern at EBKernel (Shanghai), and agent full-stack engineer at Hangzhou PeakVision AI. Leads DuoCode's architecture and the Agentic CS runtime.
Tang Chye Fong
Co-Founder, registered owner
Co-founded DUOCODE TECHNOLOGY; drives product development and the delivery of AI integration services to Malaysian SMEs, including the Agentic CS platform. Previously AI Developer Intern at PixelPro Sdn Bhd (Shah Alam), building Python web-scraping agents (Selenium, Playwright, Puppeteer) and n8n data pipelines, and evaluating RPA tooling for business intelligence; earlier, Cybersecurity Intern at Condition Zebra (M) Sdn Bhd, running penetration tests across networks, systems and applications and writing client-facing risk and remediation reports.
Jovie Lim Yee Teng
Product & Operations Assistant
Supports delivery of DuoCode's AI integration services to Malaysian SMEs. Handles product testing and feedback triage, prepares internal and client-facing documentation, and assists in client demonstrations — channelling user feedback back into product improvements.
Insights
Read Our Engineering Blog
Deep dives into AI agents, guardrails, and the technology behind DuoCode — powered by a cutting-edge editorial text engine.
Explore ArticlesFrequently Asked Questions
Common questions about the DuoCode product system
What does DuoCode actually do?
DuoCode is a multilingual, guardrail-first AI customer-service operating system for Malaysian SMEs. It answers customers from approved evidence, routes risky actions to human approval, and keeps a replayable audit trail of every run.
How does DuoCode keep an AI service loop under control?
Safety isn't a one-way pipeline. Every request runs one governed loop, and high-risk cases leave that loop for a human before anything moves. The verified outcome then becomes context for the next question.
Who is the DuoCode product for?
Malaysian D2C and SME merchants doing 200–2,000 customer conversations a month — typically on Shopee, Shopify, or WooCommerce — who can't afford the leading enterprise AI customer-service products at USD 0.99 per resolved conversation and don't trust generic chatbots not to hallucinate refund commitments. Year 1 focus: Klang Valley and Penang.
How is DuoCode priced?
Hybrid base + per-resolution. Starter is RM 299/month (200 resolutions included, RM 1.50/extra). Growth is RM 799/month (800 resolutions, RM 1.20/extra, multi-language, custom policies). Scale is RM 1,899/month (2,500 resolutions, RM 0.90/extra, unlimited stores). All tiers come with the full guardrail cockpit. Final pricing is being calibrated against the first 10 design partners.
Is DuoCode a real registered company?
Yes. DuoCode Technology is registered with the Companies Commission of Malaysia (SSM) under registration number 202603062356 (LA0087244-A) and headquartered in Klang, Selangor. The public site uses the current registered entity name consistently across legal pages, metadata, and structured data.
I'm interested as a design partner / investor — how do I get in touch?
Drop a note via the contact section above or email duocodetechu@duocodetech.com. We're actively in conversation with Malaysian D2C brands about invite-only access, and with SEA-focused angel investors about the pre-seed round. Live demos of the Agentic CS service loop and operator review surface are available on request.
Talk to DuoCode
Whether you're a SME exploring the AI agent, a partner who wants to integrate, or an investor who wants to see the demo — drop us a note and we'll be in touch within 24 hours.