The AI support rep that
actually reads your policy.

Sam answers customers in your voice, in their language, around the clock. Pulls answers from your real knowledge base and product catalog. Looks up live orders. Handles refunds inside the rules you set. Hands the conversation back to your team the moment something is outside scope, with the full context attached.

S
Sam
Customer Service Agent
Live
Hi, I ordered a black hoodie last Tuesday but it still hasn't arrived. Order #SG-4823.
Checking your shipping policy
Pulling order #SG-4823 from your store
Tracking the package live
Found it! Your hoodie shipped Wednesday and is currently at the KL sorting facility — it should arrive tomorrow by 5pm. I'll send you a live tracking link too. Anything else I can help with?
Actually, I need it by 3pm tomorrow for a gift. Can you expedite?
A 3pm guarantee is not mine to make, so I am bringing in Sarah from the team now. She can talk to the courier about the expedite and will reply right here.
Bringing in a human teammate

Handed off to a human

Handed off to Sarah from your support team with the full conversation. She can take it from here.

Ready
Real-time

Built for support teams that care what the answer actually says.

Most "AI support" tools either follow a rule tree (and break the moment a customer goes off-script) or pipe a generic LLM at your customers (and make up prices, policies, and product specs). Sam does neither. He reads from your knowledge base, your product catalog, your refund rules, and your live order data. When the answer is not in there, he says so and hands the conversation to a human with full context.

Three things, done well.

Grounded

Pulls answers from your KB and product catalog only. Refuses to invent specs, prices, or policies. If the answer is not in writing, he tells you and hands off.

Connected

Live order, shipping, and customer-history lookups through your existing tools. Handles refunds and returns inside the rules you set. Real-time, not "based on most recent training data".

Bounded

Knows when the answer is out of scope. Escalates to your team with the full transcript, the tools he tried, and the reason he stopped. Your operator picks up where Sam stopped, not from message one.

Sam reads your policies the way your best operator would.

Every reply runs through a retrieval-augmented loop. The agent searches your knowledge base for the closest matching policy, pulls in the relevant product entry from the catalog, queries live order data when the customer references an order, and only then drafts an answer. Each fact the agent uses is cited internally with a source link your operator can audit. If two policies conflict or a query falls outside the catalog, Sam stops drafting and escalates instead of guessing.

  • Knowledge base retrieval

    pgvector-backed semantic search across every policy, FAQ, and SOP you upload

  • Product catalog lookup

    live inventory, pricing, SKU specifications, and media

  • Order + shipping query

    real-time read from your storefront, courier APIs, and CRM

  • Refund + return automation

    executes inside the rules you set, never outside

  • Brand voice control

    tone, formality, and signature line set per tenant; few-shot examples from your real replies

  • Language match

    responds in the customer's language and register — English, Bahasa Melayu, Mandarin, Singlish, mixed

A real-world resolution, in under 90 seconds.

A KL-based e-commerce shopper messages at 23:14 about a parcel that has not shipped. Sam handles it before the operator wakes up. Here is the conversation, in full.

23:14

My order #SO-44218 should have shipped 2 days ago. What is happening?

Sam · 23:14

Looking that up now. One moment.
Order lookup · #SO-44218

Sam · 23:14

Got it. Your order is at the KL sorting hub. PosLaju had a 48h delay on Tuesday across the Klang Valley. Your parcel ships first thing tomorrow morning and is on track to arrive by 5pm. Here is the live tracking link: [link]
Courier API · PosLaju tracking

23:15

Can I get a partial refund for the delay?

Sam · 23:15

Checking the refund policy on this order.
KB · Refund policy v2.4

Sam · 23:15

Our policy covers a 10% store credit on parcels delayed by 48 hours or more once the parcel ships. I have queued that credit on your account, RM12.40, valid for 12 months. It will appear under your store credits the moment the parcel scans out tomorrow morning.
Refund action · 10% store credit

23:16

Thanks Sam. That works.

Resolution time: 91 seconds. Operator review the next morning: 30 seconds (approve the credit action that already fired). The customer was treated like a regular, not a ticket.

Most "AI support" is one of three things. Sam is the fourth.

FeatureRule-tree chatbotsGeneric LLM chatbotsEnterprise help-desk botsSam
Off-script reply handlingBreaksHallucinatesRoutes to humanReasons + responds
Policy groundingKeyword matchPretrained guessesAdd-on moduleNative, per-tenant
Live order lookupNoNoLimitedReal-time, every reply
Refund execution inside rulesNoNoWorkflow nodeFirst-class action
Bahasa Melayu / SinglishLimitedEnglish-firstAdd-onNative
Pricing modelRM-friendlyUSD seat-basedEnterprise contractRM-friendly

Sam earns your trust before he sends without you.

Sam does not auto-send the day you onboard. Every reply starts in draft mode, routed to an operator for approval. As your team marks his drafts as accept-verbatim, his confidence track per channel + per topic grows. When the accept rate crosses a threshold you set, he graduates to supervised, then to autonomous. Compliance verticals — clinics, financial services, education — can keep him in approval mode forever, by design.

01

Draft

Every reply reviewed by your operator. Sam learns from each edit.

02

Supervised

Sam auto-sends on high-confidence replies, escalates on low. Threshold per channel.

03

Autonomous

Sam sends without review on graduated channels. Operator audits a sample.

Questions support managers ask first.

How does Sam stop hallucinating prices or policies?
Sam is not allowed to answer from his own pretraining. Every factual claim he makes is sourced from your knowledge base, product catalog, or live tool result. If the retrieval returns nothing relevant, Sam stops drafting and either asks a clarifying question or hands off to a human. There is no "best guess" code path.
What does the handoff to a human look like?
The full transcript routes to your queue with three context blocks: what Sam tried, what he could not resolve, and what the customer actually wants. The operator picks up from message N, not from message one. There is no "let me catch up on this conversation" overhead.
Can Sam handle refunds?
Yes, inside the rules you write. You define refund eligibility (time window, condition, amount, channel). Sam executes refunds and credits that fit those rules and escalates anything outside them. Every refund action is logged with the rule that authorized it.
Which channels does Sam work on?
WhatsApp (Baileys and Meta Business API), web chat, and email today. Telegram, Instagram DM, and Facebook Messenger are on the roadmap. Channels are first-class — Sam respects each channel's rate limits, templates, and ToS.
How does Sam handle Bahasa Melayu or Singlish?
Native. Sam mirrors the customer's language and register. Same agent handles a formal Bahasa Melayu enquiry from a Penang grandmother and a Manglish "boss, where my order la" from KL on the same channel, and shifts tone accordingly. You do not need a separate agent per language.
What if Sam gets a question that needs human judgement?
He escalates. Sam knows the difference between "I do not know the answer" (escalate) and "the answer requires discretion" (escalate with a recommendation). Operator sees both the question and Sam's reasoning, so the human decision is informed, not from scratch.

A 24/7 support team that costs less than one hire.

Sam onboards in a day. Your knowledge base, your policies, your tone — captured, grounded, and live on WhatsApp by tomorrow.