Meet Spark AI.
The AI that thinks in Manglish.
Spark AI is the custom-built intelligence layer that powers every WABI workspace. It drafts every reply, learns from every correction, and writes your weekly business intelligence report — all without asking you to build a single flow.
What makes Spark AI different
Six pillars, one product
Speaks Malaysian, not just English
Bahasa Malaysia, English, Mandarin, and the Manglish in between. Customer types 'boss harga berapa ya?' — Spark AI replies in the same register, same warmth, same code-switching. The international competitors treat Bahasa as a translation step. We treat it as the default.
- Native BM / EN / 中文 support
- Code-switches mid-sentence
- Honorifics + register-matching (bro, kak, tuan, sir)
Trained on your business, not generic data
No flow builder, no decision trees, no scripts to maintain. Spark AI reads your knowledge base, your business context, and every correction your team has ever made. It's a closed loop: your team teaches it, and only your workspace benefits.
- Per-tenant learning — never shared across workspaces
- Corrections become knowledge-base drafts automatically
- Approve once, remember forever
Confidence-gated replies
Every Spark AI draft comes with a confidence score. Set the threshold (we recommend 75). Above it, the reply ships in seconds. Below it, the conversation cleanly lands in the escalation queue with a draft already prepared — your agent only edits and approves.
- Per-tenant confidence threshold (0–100)
- Full conversation context handed over to the agent
- Agent corrections feed back into Spark AI's training loop
Escalates on the words that matter
Some questions should never be answered by AI. Words like 'lawyer', 'complaint', 'refund', 'tak puas hati', or 'cancel' skip Spark AI entirely and immediately page a human. You define the trigger list. We never argue with an angry customer.
- Keyword pre-gate — skips LLM call entirely
- Tenant-configurable trigger list
- Saves cost AND avoids the worst kind of AI failure
Your AI, your model
Choose GPT-4o mini for blazing speed and low cost on the Starter plan. Upgrade to GPT-4o or Claude 3.5 Sonnet for nuanced reasoning on Growth. Switch to Claude Opus on Scale. We never lock you into one vendor's model — your business shouldn't be hostage to anyone's pricing changes.
- OpenAI: GPT-4o / GPT-4o mini
- Anthropic: Claude Sonnet / Opus
- Per-tenant model choice for replies AND learning
Singapore-resident, PDPA-aligned by design
Spark AI's data lives in AWS Singapore (ap-southeast-1). Never in the US, never in the EU, never anywhere your customer would object to. The only cross-border movement is the transient LLM call itself, with zero-retention contracts at the provider.
- Data at rest: Singapore only
- Zero-retention with OpenAI + Anthropic
- Operated by Tek Asia Sdn Bhd, a Malaysian-registered entity
Loved by Malaysian SMEs
What customers say about Spark AI
Real businesses, real Spark AI outcomes — across e-commerce, F&B, services, and retail.
“Spark AI replies in Bahasa, English, and switches mid-sentence when our customers type Manglish. DM response time dropped from 2 hours to 8 seconds. The international tools we tried before couldn't even handle 'kakak boleh refund ke?' properly.”
“Eight outlets, one WhatsApp number. Spark AI handles 84% of reservation enquiries automatically. My team only sees the special requests now — birthday cakes, allergies, big groups. That's the work humans should be doing.”
“What sold me was the keyword escalation. Any time a customer types 'allergic' or 'refund' or 'complaint', Spark AI immediately pauses and pages me. I never have to worry about the AI saying something wrong on sensitive things.”
“One DM says 'Tuan, ada iPhone 15 case ke?' — Spark AI replies formally. Next DM says '16 pro max stock ada bro' — Spark AI replies in the same casual register. It mirrors however the customer types. We didn't have to configure anything.”
“I correct Spark AI's drafts maybe 3-4 times a week. By next month, it doesn't make those mistakes anymore. It's literally learning my product line week by week — I've never seen any AI do this for our specific policies.”
“Switched from WATI six months ago. The PDPA story sealed it — our data sits in Singapore, not the US, and the DPA was already signed by Tek Asia. My compliance officer hasn't found a single thing to argue with.”
vs. The alternative
Why “just add ChatGPT” isn't enough
The shortcut everyone tries first — and what breaks at scale.
| Spark AI inside WABI | DIY ChatGPT integration | |
|---|---|---|
| Manglish + code-switching | Native — same model, same conversation | English only; BM via clunky translation step |
| Learns from your team's edits | Automatic — corrections become knowledge base drafts | You manually re-prompt or fine-tune |
| Confidence-gated handover | Built in — threshold per tenant | DIY — you write the gating logic |
| Keyword-based human escalation | Built in — skips LLM entirely on flagged words | DIY — and you pay for every LLM call regardless |
| Data residency in Singapore | Yes — AWS ap-southeast-1 | OpenAI in the US, no residency control |
| PDPA-aligned controller relationship | Tek Asia Sdn Bhd, signed DPA | You become the controller, on your own |
| Cost guardrails | Per-tenant daily spend cap, pre-gates | Surprise OpenAI bill at month end |
| Setup time | Connect WhatsApp + paste knowledge base → live tonight | Weeks of integration + flow design |
Built-in guardrails
Safety isn't a feature. It's the default.
Cross-tenant isolation
Postgres row-level security enforces tenant scoping at the database layer. There is no SQL path by which Tenant A can read Tenant B's data — not by bug, not by query mistake.
Daily LLM spend cap
Each tenant has a hard daily spend ceiling. Hit the cap, and Spark AI gracefully falls back to keyword-routing and human escalation until midnight. You'll never get a runaway-cost incident.
Zero-retention LLM contracts
OpenAI and Anthropic process your messages and return the answer — they don't keep, log, or train on your content. We hold them to this contractually and verify quarterly.
Try Spark AI
Connect your WhatsApp number. See it work tonight.
14-day free trial. No credit card. Spark AI handles your first inbound message within minutes of signup.