Frequently Asked Questions

Chung Tair Ltd. was founded in 2025 and is headquartered in Kaohsiung, Taiwan, focusing on AI customer service automation. We build multi-channel AI customer service for Asian SMEs, integrating LINE, Facebook Messenger, and Instagram plus an embeddable website Web Widget. Powered by RAG (Retrieval-Augmented Generation), HyDE, and Hybrid Search, the AI answers accurately from each business's own knowledge base, in Traditional Chinese, English, and Japanese.

In plain terms, these technologies serve three goals: accurate answers, no making things up, and automatic human handoff when stuck. Chung Tair AI is built on a three-layer reply architecture (Fixed Reply → RAG → Fallback), combined with several advanced techniques: HyDE generates a hypothetical answer before vector search; Hybrid Search runs vector and keyword search in parallel, merged via Reciprocal Rank Fusion (RRF); Multi-Query Retrieval searches the rewritten and original questions together; Query Rewriting restores pronouns and context; Contradiction Detection flags conflicts on write; Topic Summary Auto-Compilation auto-clusters and summarizes. The vector store is PostgreSQL + pgvector (1536-dim embeddings, ivfflat index), with a multi-knowledge-base architecture (private customer KB + shared system KB) searched in parallel.

Chung Tair AI supports three social platforms — LINE, Facebook Messenger, Instagram — plus a website Web Widget you embed on your own site (paste one <script> before </body>). With Rich Media Reply, the AI can return text, images, and URLs together, natively in each channel. Deploy once and run across every supported channel with a single shared knowledge base.

Chung Tair AI natively supports Traditional Chinese, English, and Japanese, automatically replying in the language the customer uses. It uses OpenAI multilingual models, tuned for common Asian-market contexts (business Japanese, colloquial Traditional Chinese, English support phrasing) for consistent cross-language quality.

In plain terms, RAG makes the AI answer only from your knowledge base instead of inventing answers — the 'no bluffing' that customer service needs most. RAG (Retrieval-Augmented Generation) customer service has the AI look things up in your knowledge base first, then generate an answer from what it retrieved. The key difference from keyword auto-reply: a keyword bot matches the customer's text against preset keywords and returns a fixed canned message, so it fails the moment the customer rephrases. RAG instead understands the meaning of the question, finds the most relevant content in your knowledge base, and composes a natural-language answer — so it understands however the customer asks, and answers stay grounded in your knowledge base rather than made up. Chung Tair's RAG uses a three-layer architecture (Fixed Reply → RAG → Fallback) and combines HyDE, Hybrid Search and more to raise the hit rate.

In plain terms, HyDE helps the AI better grasp what the customer is actually asking, so answers are more accurate. HyDE (Hypothetical Document Embeddings) is a technique that improves vector-search accuracy. Normally you match the customer's (short, colloquial) question directly against the knowledge base, but the wording gap with formal KB entries makes hits unreliable. HyDE instead first has the LLM generate a hypothetical ideal answer to the question, then runs vector search with that hypothetical answer — because its wording and context are closer to the real answers in the knowledge base, hit rate improves. Chung Tair applies HyDE to article-type long entries to strengthen semantic matching on longer passages.

The biggest differences are whether it understands meaning and where the answer comes from. Traditional chatbots are mostly keyword/button flows - the customer must follow preset options and gets stuck when phrasing things naturally; modern AI customer service uses an LLM to understand meaning and RAG (retrieval-augmented generation) to find answers in your knowledge base, so it understands however the customer asks and stays grounded. In short: a chatbot follows a script, AI customer service understands and then answers. Chung Tair is the latter, with a three-layer RAG architecture, human handoff, and multi-channel support (LINE / FB / IG / website).

Yes - good multi-channel AI customer service should let you build the knowledge base once and share it across all channels. You should not maintain separate answers per platform; wherever the customer comes from, the AI replies from the same knowledge base, and the dashboard shows all conversations in one place. Chung Tair supports four channels at once - LINE, Facebook Messenger, Instagram, and a website Web Widget: FB/IG via one-click OAuth, LINE by pasting a token (or using our setup service), and the website with one snippet. Setting changes apply instantly from the dashboard, with no need to reconfigure each channel.

Chung Tair AI uses fully automatic human handoff with four trigger types: (1) intent match (the LLM detects the customer intent matches a merchant-defined handoff scenario); (2) AI self-assessment (a built-in five-step self-check); (3) consecutive fallback (default 2, configurable 1–10); (4) passphrase trigger. On handoff the LLM generates a summary for the agent; an AI Pause Timer automatically resumes the AI after N minutes. Inbound images also support a three-way policy (handoff / ask for text / ignore).

Chung Tair AI has built-in Contradiction Detection: when a new entry is added or updated, the system first screens similar entries by vector similarity ≥ 0.6, then asks the LLM to compare numeric conflicts (prices, quantities) and policy conflicts (return rules), flagging risky entries for admin review. If the LLM fails, graceful degradation does not block the write. Topic Summary Auto-Compilation also auto-clusters and generates unified summaries (summary entries get a 1.4× boost at the RRF stage), keeping answers consistent and correct.

Chung Tair minimizes room to make things up with a three-layer reply architecture: Fixed Reply -> RAG -> Fallback. The first layer is merchant-set fixed answers (key facts like price or opening hours are matched directly, not AI-generated); only the second layer is RAG, where the AI answers strictly from content retrieved from your knowledge base, not from thin air; the third layer is Fallback - when nothing relevant is found, the system does not bluff but gives a safe reply or hands off to a human per your settings. Contradiction detection also blocks conflicting info as you build the KB, and knowledge-gap tracking logs unanswered questions by frequency for you to fill. So accuracy ultimately depends on how complete your knowledge base is - with thorough material, the AI answers accurately and with grounding.

Good AI customer service must be able to hand off to a human - pure AI cannot handle every situation, and lacking this actually frustrates customers. What matters is when it hands off and how smooth that is. Chung Tair uses fully automatic human handoff with four triggers: the AI detects the customer's intent matches a handoff scenario you set, the AI self-assesses that it answered poorly, consecutive no-answers hit a threshold, or the customer types a passphrase. On handoff, the AI generates a conversation summary for the agent, and a Pause Timer lets the AI resume automatically after N minutes without talking over the human.

Whether it answers wrong comes down to whether it can stay silent when unsure. A common problem is AI that bluffs when it does not know (hallucination) - that is the real risk. The safe design is: when no sufficiently relevant material is found, do not guess but give a safe reply or hand off to a human. Chung Tair achieves this with a three-layer architecture (Fixed Reply -> RAG -> Fallback) - key facts use merchant-set fixed answers, general questions use RAG strictly grounded in the knowledge base, and when nothing is found it falls back rather than bluffing; contradiction detection also blocks conflicting info as you build the KB. Accurate answers depend on a complete knowledge base.

What matters is how well the fallback is designed when it cannot answer. A poor system replies randomly or keeps spamming canned messages; a good one tries multiple ways to understand and, if it truly cannot, gives a safe reply or hands off to a human. Chung Tair first uses Query Rewriting to restore pronouns and context and Multi-Query to search the rephrased question, doing its best to understand; compound questions are auto-split and answered separately. If still no relevant content is found, it falls back (safe reply or handoff per your settings) and logs the unanswered question in knowledge-gap tracking, ranked by frequency to prompt you to fill it - so every miss becomes an improvement clue.

Chung Tair AI is a multi-channel customer service system integrating LINE, Facebook Messenger, Instagram, and a website Web Widget, built on a three-layer RAG architecture. Key features include —
Channels: one-click OAuth for all four channels; paste one <script> for the Widget; dashboard changes apply instantly without re-pasting.
Retrieval: three-layer architecture (Fixed → RAG → Fallback) to prevent hallucination, HyDE, Hybrid Search + RRF, Multi-Query, Query Rewriting with compound-question detection, multi-KB.
Reply quality: mixed text + image + URL replies (each card relevance-checked), compound-question splitting (one message asking several things is answered point by point).
Operations & handoff: four-trigger automatic human handoff + Pause Timer + auto handoff summary, inbound-image policy, knowledge-gap auto-tracking ranked by frequency.
Knowledge base: document direct-read import (drag in whole PDF/Word/Excel files), contradiction detection.

You do not train a model like an engineer - all you do is feed it material. Modern RAG-based AI customer service does not require machine-learning knowledge; its smarts come from the knowledge base you provide, not from tuning parameters. You just organize common Q&As and product info into the dashboard, and the AI answers from that. Chung Tair makes this step easier: drag existing PDF / Word / Excel files into the dashboard to auto-split and build the KB, with no line-by-line entry; afterward, knowledge-gap tracking tells you which answers are still missing so you can fill them gradually.

Yes - this is what saves time in modern AI customer service: you do not enter document content line by line. Upload your existing product info, price lists, and FAQ documents, and the system reads and splits them into knowledge entries automatically. Chung Tair supports dragging in whole PDF / Word / Excel files to auto-build the knowledge base, saving re-typing; during the build, contradiction detection flags new material that conflicts with existing entries (such as inconsistent prices) for your review, preventing the AI from receiving contradictory information.

A great fit - Chung Tair is built exactly for Asian SMEs of this size. A ~10-person ecommerce team usually has no dedicated large support center, yet gets flooded daily with repetitive questions (shipping, returns, order status, specs). Once connected to your LINE / FB / IG, the AI answers these high-frequency questions from your knowledge base, leaving humans to handle judgment cases; when the AI falls short or a customer asks, it auto-hands off to a human with a conversation summary so nothing slips through. Knowledge-gap tracking also lists questions customers ask often but you have not answered yet, helping you keep improving. At a fixed price from NT$2,000/month with no contract, the cost pressure on a small team is low.

Any industry where customers ask via LINE / FB / IG DMs and the questions are highly repetitive. Common ones include ecommerce and online retail (shipping, returns, order tracking), appointment-based services (salons, clinics, tutoring - hours and bookings), local stores and dining (menu, address, hours, reservations), education and courses (content, fees, enrollment), and any SME with a stable FAQ but not enough hands to reply. As long as you organize your common Q&As into the knowledge base, the AI replies 24/7 on your behalf, handing off to a human for cases that need judgment.

No engineer and no coding - and both methods go live in about 5 minutes. Method 1 (DIY): follow our detailed step-by-step guide - add a LINE bot in the Chung Tair dashboard to get a dedicated Webhook URL, enable the Messaging API in LINE to get your Channel Access Token and Channel Secret, paste them back into Chung Tair, then paste the Webhook URL into LINE and enable it. Method 2 (we set it up): if you would rather not, we offer a setup service and connect it for you, also in about 5 minutes. FB and IG use one-click OAuth, which is even faster.

Yes - copy one <script> snippet before </body> on your site and you are done. Once embedded, an AI chat window appears at the bottom-right of your site, using the same knowledge base as LINE / FB / IG with no extra setup. Later changes (welcome message, handoff scenarios) are made in the dashboard and apply instantly - no need to re-paste the code. This website Web Widget is the demo itself - the chat button at the bottom-right of this very site is the same thing.

Connecting a channel takes about 5 minutes; the real time goes into organizing your knowledge base - but the barrier is low. You need no engineer and no coding. What to prepare is your FAQ material: drag your existing PDF / Word / Excel files (product info, price lists, FAQ docs) into the dashboard and the system auto-splits and builds the knowledge base - no manual entry line by line. The more complete the material, the more accurately the AI answers from day one. Channel connection (LINE / FB / IG / website Widget) can be DIY or done by our setup service. Overall, with your material ready, you can go live the same day.

Yes. Via LINE's Messaging API, a third-party AI customer service system can connect to your LINE Official Account, letting AI replace or augment the built-in keyword auto-reply. LINE Official Account's built-in auto-reply only does keyword triggers; to make it understand meaning and answer from data, you connect an AI service like Chung Tair. Once connected, when customers message your LINE OA the AI replies from your knowledge base and hands off to a human when it cannot answer. Connecting needs no engineer - follow the illustrated guide to paste a token, or use our setup service, in about 5 minutes.

Yes - no engineer and no coding. Modern AI customer service platforms package the technical parts so you only operate in the dashboard. With Chung Tair, connecting LINE has two paths: follow the detailed illustrated guide to paste your Channel Access Token and Webhook URL yourself (about 5 minutes), or use Chung Tair's setup service and we connect it for you (also about 5 minutes). After that, all you do is organize the knowledge base (drag in PDF / Word / Excel to auto-build it) - no code at all.

For SMEs whose staff are eaten up by repetitive questions every day, it usually pays off. The key is: how many of your support questions are high-frequency repeats? If shipping, returns, hours, and specs make up most of them, AI customer service handles those automatically around the clock, freeing staff for cases that need judgment - trading a fixed monthly fee for a lot of repetitive labor time. Chung Tair starts at NT$2,000/month with no contract and a 14-day free trial, so you can measure during the trial how many of your questions the AI can take over before deciding if it is worth it.

Yes - ecommerce is one of the scenarios where AI customer service pays off fastest. Ecommerce questions cluster heavily around shipping, delivery time, returns, specs, and order status - highly repetitive with relatively fixed answers, perfect for AI. The approach: organize these common Q&As and product info into the knowledge base; once the AI connects to your LINE / FB / IG / website it replies 24/7, handing off to a human for order lookups or judgment cases. Chung Tair auto-splits compound questions (a customer asking about shipping + delivery + returns at once is answered point by point) and uses knowledge-gap tracking to list frequently asked but unanswered questions for you to fill.

Yes - this is exactly the core pain AI customer service solves. Repetitive questions pile up because they are high-volume and concentrated (the same shipping or returns question asked hundreds of times). Hand those high-frequency Q&As to the AI and your staff are left with only the few cases needing judgment. The key is that the AI must actually answer correctly - Chung Tair answers from your knowledge base via RAG (not by bluffing), hands off to a human when unsure, and logs unanswered questions for you to fill. In practice, building the 20-30 most-asked questions into the knowledge base blocks most repetitive inquiries.

Very suitable. Appointment-based businesses - salons, clinics, tutoring centers, studios - are most often asked about hours, address, prices, available slots, and services, all fixed answers best handled automatically by AI. When a customer DMs at night asking about openings tomorrow or pricing, the AI replies instantly instead of making them wait until you are open, reducing missed customers. For actual scheduling or special needs, it hands off to a human. Chung Tair works once connected to LINE / FB / IG; you just organize your business info and common Q&As into the knowledge base.

Usually you just copy one snippet onto your site - no need to change your site's architecture. Most AI customer services provide an embed <script> you paste before </body>, and a chat window appears at the bottom-right. Chung Tair's website Web Widget works this way: copy one <script> into your site, using the same knowledge base as LINE / FB / IG with no reconfiguration; later changes (welcome message, handoff scenarios) apply instantly from the dashboard. The chat button at the bottom-right of this very site is that same widget.

Fixed monthly pricing, no contract. Three plans: Basic NT$2,000, Advanced NT$3,500 (most popular), and Pro NT$4,500. They differ by the number of AI bots (2 / 3 / 4) and monthly messages (2,000 / 4,000 / 6,000); knowledge-base entries are unlimited on every plan. New users get a 14-day free trial with no credit card required.

Plan	Monthly	AI bots	Messages/month
Basic	NT$2,000 / mo	2	2,000
Advanced (most popular)	NT$3,500 / mo	3	4,000
Pro	NT$4,500 / mo	4	6,000

Yes — a 14-day free trial. No credit card required, no charges, and full feature parity with paid plans; cancel anytime during the trial at no cost. One trial per applicant email.

No contract; cancel anytime. Service continues to the end of the current billing period, with no extra charge.

Each inbound customer message (including dashboard test-window messages) counts as 1; AI replies and knowledge-base training are not counted. For example, one question answered by three AI messages still counts as 1.

Yes. After each successful charge, the system automatically issues a cloud e-invoice via Cetustek and emails it to your designated address. Businesses receive a triplicate cloud invoice (please provide your tax ID and invoice title); individuals receive a duplicate cloud invoice (mobile barcode or email carrier). Cetustek is an established Taiwan e-invoice provider, ISO 27001 certified.

Chung Tair never touches your card data directly. All transactions are processed by TapPay (Cherri Tech) with industry-standard encryption, compliant with Visa, Mastercard, and JCB rules. Our servers store only transaction results and authorization codes, never full card numbers. Every charge passes 3D Secure verification.

LINE AI customer service pricing mainly depends on how many accounts/bots you connect, how many messages per month, and whether there is a contract - prices vary widely. Some charge by usage (more messages cost more, bills harder to predict), others a fixed monthly fee (easier to budget). Chung Tair uses a fixed monthly fee with no contract: Basic NT$2,000, Advanced NT$3,500, Pro NT$4,500, differing by AI bot count (2/3/4) and monthly messages (2,000/4,000/6,000), with unlimited knowledge-base entries on every plan, plus a 14-day free trial with no credit card.

AI customer service prices span a wide range - from a few hundred to over ten thousand NT dollars a month - mainly differing by pricing model and feature completeness. Two common models: usage-based (charged per message or per 'resolution', where bills climb and get unpredictable at volume) and fixed monthly (predictable budgeting). When choosing, do not look at the monthly number alone - check whether human handoff, knowledge base, and multi-channel are built in, since many cheap plans charge extra for these. Chung Tair is a fixed fee from NT$2,000/month with no contract, bundling human handoff, unlimited knowledge base, and multi-channel, with a 14-day free trial.

Chung Tair is LINE-native multi-channel AI customer service built for Asian SMEs, at a fixed monthly price from NT$2,000, bundling human handoff, knowledge-base contradiction detection, mixed text/image/URL replies, compound-question splitting, and document direct-read import — things most tools in the same price range do not do, while global vendors cost several times more and do not treat LINE as a native channel.
Chatbase starts around US$40/month but has no built-in human handoff (it requires an external Zendesk/Intercom); Intercom charges US$0.99 per resolution plus seat fees plus Copilot add-ons, making bills hard to predict; Zendesk is seat-based and complex to configure. Make/n8n can send images via workflows but are complex to set up and cannot one-click connect a LINE Official Account. Other lightweight Western tools (such as Tidio) are likewise not LINE-native and usually charge extra for AI.

Comparison	Chung Tair	Chatbase	Intercom/Zendesk	Make/n8n
LINE-native one-click OA	✅ one-click	❌	❌ not first-class	❌ manual setup
Built-in human handoff	✅ custom-scenario auto	❌ needs external helpdesk	✅ but higher tiers	❌
KB contradiction alert	✅	❌	❌	❌
Mixed image + URL reply	✅ native	text-first	partial	✅ but complex setup
Compound-question splitting	✅ separate answers	❌	❌ single reply	❌
Document direct-read import	✅ PDF/Word/Excel	⚠️ char limits, manual	⚠️ mainly help-center	❌
Pricing model	fixed from NT$2,000/mo	usage credits + overage	seat + per-resolution	tool fee + heavy setup

Last updated 2026-06-01 | Basis: each vendor's / platform's public pricing and feature pages

LINE Official Account's built-in auto-reply is keyword-triggered — the customer's text must hit a keyword you preset, otherwise it returns a default canned message. Chung Tair connects to that same LINE Official Account (via the Messaging API) and uses RAG to understand meaning, answers from your knowledge base, and adds three-layer hallucination prevention, automatic human handoff, and mixed text/image/URL replies. So it is not either/or — Chung Tair upgrades your LINE OA from a keyword bot into an AI that understands natural language. You keep the same LINE Official Account; only the brain behind the replies changes.

Aspect	LINE OA built-in auto-reply	Chung Tair AI (on your LINE OA)
Trigger	literal keyword match	RAG semantic understanding, handles rephrasing
On a miss	returns a default canned message	searches KB + three-layer guard; safe reply only if nothing found
Multi-turn / context	none	auto-restores pronouns and context
Knowledge base	none; keyword by keyword	drag in PDF/Word/Excel, auto-built
Human handoff	manual switch to chat mode	custom-scenario auto + summary + Pause Timer
Image + URL	possible but tied to keywords	delivered with the answer, each card relevance-checked

Last updated 2026-06-01 | Basis: each vendor's / platform's public pricing and feature pages

Chung Tair is built for Asian SMEs with LINE as a core channel. To be honest, it is not the best choice in these cases: (1) your main market is purely Western and you do not use LINE / Facebook / Instagram — those markets center on WhatsApp and web chat, where other tools fit better; (2) your support is mainly phone/voice — Chung Tair is AI for text channels and does not handle voice calls; (3) you do not want to build any knowledge base at all — Chung Tair's accuracy comes from your KB, so with no material to learn from, results are limited; (4) you need enterprise-grade custom workflows and large-team seat management — platforms like Intercom or Zendesk suit that better. If you are an Asian SME centered on LINE that wants affordable, accurate AI answering from your own knowledge, Chung Tair is built for you.

When you find LINE's built-in auto-reply cannot handle rephrasing, only matches keywords, has no multi-turn dialogue, and cannot look things up, it is time to upgrade to AI customer service. Upgrading does not mean replacing your LINE Official Account - you connect an AI service via the Messaging API, keeping the same account and followers; only the reply brain changes from keyword matching to meaning understanding plus a knowledge base. Once connected, Chung Tair's AI understands questions via RAG, answers from your knowledge base, auto-hands off to a human when unsure, and can include images and URLs. Connecting takes about 5 minutes, DIY by guide or via our setup service.

Do not pick by monthly fee alone - use this checklist: (1) does it use RAG/semantic understanding (not pure keywords); (2) is human handoff built in (many tools require an external add-on); (3) does it support your main channels (in Asia, usually LINE-native); (4) is the knowledge base easy to build (can it import documents directly); (5) does it bluff when it cannot answer (is there fallback / hallucination prevention); (6) is pricing predictable (fixed monthly vs usage-based); (7) is there a free trial to test for real. Chung Tair covers all of these: three-layer RAG, built-in human handoff, native LINE / FB / IG / website, document direct-read KB building, Fallback hallucination prevention, fixed fee with no contract, and a 14-day free trial - testing with your own questions during the trial is the most reliable check.