Five guardrail layers stop hallucinations before they reach your customer. Native FPX, GrabPay, Boost, Shopee, J&T, and Pos Laju integrations. Pricing that doesn't punish you for being a small business.

DuoCode operator cockpit — RM500 refund queued at L5, waiting on one-click approval
The leading enterprise AI customer-service products list at USD 0.99 per resolved conversation — built for the Fortune 500. For a Malaysian D2C brand doing 1,000 tickets a month, that is RM 4,500/month minimum, billed in USD, with English-first defaults and no native FPX/GrabPay support.
Generic chatbots hallucinate refund commitments, return windows, and stock numbers. For a small brand, one over-promised refund is a chargeback and a 1-star review — recent public industry incidents have shown those commitments can be enforceable against the operator.
DuoCode is the agent built for the SME that wants AI without losing sleep. Multilingual on day one. Guardrailed at five layers. Priced like a SME tool, not an enterprise contract.
Every other AI customer-service product is a single LLM call wrapped in a prompt. DuoCode is a pipeline of five enforcement layers, each catching a different class of failure — already running end-to-end in our operator cockpit.
Catches prompt-injection patterns and forbidden language ("ignore previous instructions", unauthorized guarantees) before any LLM round-trip — saving tokens and blocking the attack at the cheapest layer.
Every factual claim in the agent's draft must trace back to a tool result or a cited policy chunk. Unsourced claims are blocked or regenerated.
Every tool call (refund, exchange, address change, ticket) is validated against a strict Zod schema and a cross-call identity check before execution.
A second LLM pass audits the draft for unauthorized commitments, guarantee language, and tonal misalignment — catching subtle hallucinations the regex layer misses.
Actions above merchant-defined limits (refund above RM200, ≥3 refunds in 30 days, address change after shipment) are queued for the merchant to approve in one click. The tool call never auto-executes.

L5 — Human approval dialog with refund_id + audit_id receipt

Final response — every claim cites a policy chunk or tool result
Numbers from our internal evaluation harness — 30 YAML test cases (10 happy · 8 guardrail · 7 adversarial · 5 policy QA) against 3 configurations.
Safety rate (full guardrail stack)
p95 latency cost of guardrails
Eval cases passed
Independent enforcement layers
| Capability | Enterprise AI CS | Local rule-based bots | DuoCode |
|---|---|---|---|
| Pricing for SME | Out of reach | Affordable | Affordable |
| Hallucination guardrails | Partial | Rule-based only | 5-layer pipeline |
| Native Malay / Chinese | |||
| FPX / GrabPay / J&T | Partial | ||
| Human-in-the-loop UX | Limited | — | Core to product |
| Setup time for SME | Weeks | Days | Hours |
Hybrid base + per-resolution model. You pay more only when the agent is actually working for you — a quiet month is a quiet bill.
Up to 200 AI-resolved conversations · RM 1.50 per additional · single language · single store
Up to 800 resolutions · RM 1.20 per additional · multi-language · up to 3 stores · custom policies · full guardrail cockpit
Up to 2,500 resolutions · RM 0.90 per additional · unlimited stores · priority support · custom integrations
Final pricing calibrated against feedback from the first 10 design partners. We'd rather get this right than push a number.
We are raising RM 500,000 to take DuoCode from prototype to first 30–50 paying Malaysian SME customers over an 18-month runway.
We are looking for investors who understand SME SaaS economics, have a network in the Malaysian/SEA D2C ecosystem, and are comfortable with a technical, product-led team and a deliberately narrow Malaysian beachhead. In return: exposure to the localized AI agent layer for one of the fastest-growing e-commerce regions in the world.
duocodetechu@duocodetech.com · +60 14-739-8281 · Klang, Selangor, Malaysia