Website support agent
L1 deflection done right: answers product, pricing, and how-to questions from your own docs and site — and hands the tricky ones to a human with full context. Like Fin, but yours.
- Grounded answers
- Human handoff
- 24/7
AI Automation · Conversational AI Singapore
Chatbots that answer.
Not deflect.
Singapore B2B buyers won't tolerate a decision-tree bot that loops them back to a contact form. As an AI chatbot agency, we build grounded conversational agents that actually answer — on your website, your WhatsApp, your support desk. The assistant in the corner of this page? We built that one.
- First bot live 2–3 weeks
- Grounded in Your docs, not guesses
- Languages EN · 中文 · Malay · Arabic
01 Proof, not slides
The bot in the corner is ours.
Most "AI chatbot agencies" demo someone else's tool. The assistant on this site — "Fin" — was designed and built by us, and it's a fair sample of what you'd get. Open it and ask it something.
Claude-poweredAnthropic, swappable
GroundedAnswers from our content
BilingualEnglish & Arabic, live
Routes to humansBooks a call, no dead ends
Our own infraNo third-party SaaS
Logged & reviewedEvery chat, for tuning
02 What we build
Six kinds of bot
worth shipping.
Same grounded brain underneath — different jobs, different front doors.
Lead-qualification bot
Qualifies visitors against your ICP, captures intent, and books the meeting straight into your calendar — so sales talks to people who are actually ready.
- ICP scoring
- Calendar booking
- CRM sync
Internal knowledge copilot
Your team asks policies, SOPs, product specs, and contract terms in plain language — answered from your wiki and drive, on Slack or Teams, with permissions respected.
- Slack / Teams
- Permission-aware
- Cited sources
WhatsApp & multichannel
SEA runs on WhatsApp. We deploy the same agent to WhatsApp Business, web, and your help desk (Intercom, Zendesk, Freshdesk) — one brain, every channel.
- WhatsApp Business
- Help-desk native
- Unified logs
Multilingual & Singlish-aware
English, Mandarin, Malay, Bahasa, and Arabic (RTL) — tolerant of Singlish and code-switching, so it sounds like it belongs in the region, not translated into it.
- 5+ languages
- RTL support
- Code-switching
Guardrails & evals
Retrieval-grounded, with refusal behavior, PII handling, and an eval suite we agree on up front — so you can prove the bot is accurate, not just hope it is.
- Eval suite
- PII / PDPA
- Refusal logic
03 How we ship
Ground it first.
Then let it talk.
Define
Which questions it must nail, which channels, which languages, and the accuracy bar. We map the content it'll answer from.
Ingest
We pull in your docs, site, and knowledge base, build the retrieval layer, and wire guardrails, refusals, and human handoff.
Test
We run it against a real question set, tune until it clears the bar, and shadow it on live traffic before it answers anyone for real.
Ship & own
Live on your channels, conversations logged to your dashboard, and the whole thing handed over — prompts, evals, code — for your team to run.
04 Stack
Best model for the job. Your infra underneath.
We pick the model on accuracy, latency, cost, and your data-sensitivity — not a kickback partnership. It all runs in your accounts.
Anthropic ClaudeDefault for grounded chat
OpenAI GPTWhere it fits
Llama · MistralSovereign / on-prem
RAG / vector DBpgvector · Pinecone
WhatsApp Business APIChannel delivery
Your cloudAWS · GCP · Netlify
Try it, then talk to us
Ask the bot first.
Then book the call.
Open the assistant in the corner and put it through its paces — that's our work. Then tell us the conversation your team keeps having by hand, and we'll scope the bot for it. See pricing.
05 Chatbot FAQ
The questions buyers
actually ask.
How much does an AI chatbot cost in Singapore?
A grounded production chatbot usually sits at the lower end of our build range — fixed-scope, priced before we start, then an optional monthly retainer to improve it as your content and questions evolve. See our pricing breakdown, or book a scoping call for a fixed quote.
Will the chatbot hallucinate or make things up?
We ground every bot in your own content — docs, site, knowledge base — using retrieval, not the model's open memory. It answers from your material, cites where it can, and is built to say "I don't know, let me connect you to a human" rather than invent. We agree an accuracy/eval bar before launch and monitor it after.
Can it handle multiple languages and Singlish?
Yes. Our agents handle English, Mandarin, Malay, Bahasa Indonesia, and Arabic (with right-to-left support), and tolerate Singlish and code-switching. The assistant on this page already runs in both English and Arabic depending on the visitor.
Where does our data go — is it PDPA-compliant?
The bot runs on your own infrastructure and accounts. We configure data residency for embeddings and logs, run a PDPA review of the conversation data flow, and set up retention and purge rules for anything touching personal data. Nothing critical is trapped on our side.
Which channels can the chatbot run on?
Website widget, WhatsApp Business, Slack or Microsoft Teams for internal bots, and most major help desks (Intercom, Zendesk, Freshdesk). Same grounded brain, different front doors.
How fast can you launch a chatbot?
A first production chatbot is usually live in 2–3 weeks — grounded in your content, evaluated against real questions, and handed to your team to own and extend.