Limited-time offer · dedicated onboarding help

QUBBY AI · Enterprise Voice Intelligence

An AI that answers the phone,
not just one that talks.

Qubby turns every call your business handles into an AI employee that answers, dials, looks things up, and files records — picking up incoming calls, running batch outbound, transferring to a human when busy, capturing leads mid-call, and even reading your internal systems to fetch data on its own.

1,000+ enterprises adopted-70% call handling time95% appointment completion rate

01 — AI IP Phone · The Phone Brain

Put AI inside the phone — your phone's new brain

At the core of Qubby is the AI IP Phone — AI built directly into a SIP phone to become the brain of your phone system. It comes online like a standard IP handset, can dial out proactively, decide how to route incoming calls, understand the conversation, look up data, and transfer to a human. It registers to your on-premise PBX, your Cloud PBX, or connects directly to a carrier SIP Trunk; and because it's a fully compliant SIP endpoint, it travels the phone system's standard channels and won't be blocked by firewalls or security policies.

  • AI = the phone brain — fused with a SIP phone to dial out, route incoming calls, converse and look up data, and transfer to a human, all in one endpoint.
  • Three ways to connect — existing on-premise PBX (SDX / iPECS), Cloud PBX, or carrier SIP Trunk, all supported with no hardware changes.
  • Never blocked by security — comes online as a standard, compliant SIP endpoint on existing channels; no special firewall rules, never stopped by security policies.
  • Nearby data centers — multiple regions including Taiwan and Thailand; the AI runs in the data center closest to you for low latency and stable calls.
3 WAYS TO CONNECT
Existing on-premise PBX
SDX · iPECS · SIP PBX
Qubby AI
Phone brain · SIP endpoint
Cloud PBX
Cloud PBX
Carrier SIP Trunk
Direct carrier trunk connection
Proactive outboundIncoming routing decisionConverse · Look up dataTransfer to human
BrainAI fused with a SIP phone → one endpoint can dial out, route incoming calls, converse, and transfer to a human.SecureA standard, compliant SIP endpoint on existing channels → no special firewall rules, never blocked by security.

02 — AI Smart Receptionist · How It Works

Once a call connects, how it finds the right person

The smart receptionist isn't a touch-tone menu; it's a real-time voice conversation: the caller simply says who they're looking for, the AI transcribes the speech instantly and runs a fuzzy match against your directory, locks onto an extension, and transfers through your existing PBX. Below is the four-stage journey a call actually takes through the system.

  • No extension needed — matching uses Chinese word segmentation + keywords, so saying "find sales" or "a warranty issue" still reaches the right person in "Sales / Repair."
  • No wrong transfers on shared names — two people with the same name but different extensions are distinguished by department / title / alias / keyword to pick the correct one.
  • Fallback when not found — if no person is found, it takes a message or routes to a default extension instead of leaving the caller hanging.
  • Bulk directory import — Excel / Word / PDF / photos all have their fields recognized, including department, keywords, and aliases.
CALL PIPELINE

The four stages of an incoming call in the system

ConnectedPBX → AI extension
Live transcriptionVoice → text
Directory match"Find sales" → Manager Wang 302
PBX transferConnect to extension 302
02 Live transcription: transcribes as it listens, no need to wait for the full sentence.03 Fuzzy match: Chinese word segmentation + IDF weighting, tolerant of colloquialisms and homophones.04 Transfer: routes through your existing PBX internal extension, AI drops off the line.

03 — Human Transfer · Two Methods

Two ways to reach a human — does the line stay busy?

Continuing on: once the receptionist finds the right person, how does it hand the call over? Two methods — both work for incoming phone calls and web voice service. Conference bridging (ConfBridge) pulls both parties into one call with the AI staying on the line; line transfer (REFER) asks the PBX to connect the caller straight to the extension while the AI drops out. The difference is call resources — when lines are limited, REFER frees the line once the transfer completes.

  • Bridging — the AI can record the whole call and step back in when needed; the trade-off is holding a line.
  • Transfer (REFER) — after handing off to the extension the AI fully drops out, holding no line, ideal for PBXs with limited lines.
  • Auto-selected by PBX capability — configurable in the backend; it also plays a ringback and confirms the extension is available before transferring.
TRANSFER MODES

A Conference bridge

CONFBRIDGE
Incoming
AI
Extension
AI stays on line · holds 1 line

B Line transfer

REFER
Incoming
AI
Extension
AI drops out · frees line, 0 held
See diagramLeft: AI stays lit (line held); right: AI fades out (drops off after transfer), caller and extension talk directly.

04 — AI Smart Service · How It Works

Its answers are "looked up" from your own documents

This isn't just a chatbot. After you upload product manuals, PDFs, or your website, Qubby splits the content into chunks and converts them into vectors to build a knowledge base. When a customer asks a question, it uses semantic search (meaning, not keyword matching) to find the most relevant passage with a confidence score; it answers from that passage only when confident enough, and says it doesn't know and transfers to a human when confidence is low — so it never makes things up.

  • Meaning over wording — asking "what do I do if it's broken" still matches "repair request process."
  • Confidence threshold — every match has a score; below the threshold it would rather say it doesn't know than guess.
  • Manageable data — multi-file knowledge bases can be toggled file by file, updated to new versions, and traced to which document answered.
  • One setup, many channels — the same service agent works on the phone, web voice, and web text chat at once.
RETRIEVAL → ANSWER
Your documents
product-manual.pdfwarranty-policy.docxWebsite FAQ
Chunk + vectorize

Customer asks: "My machine is broken, how do I handle it?"

① Warranty policy · Repair request processRelevance 0.91 ✓
② Product manual · Installation guide0.42
③ Website FAQ · Returns0.31
FoundTake the most relevant passage (0.91) → answer from that passage, with the source cited.Not foundTop score still below threshold → reply "let me transfer you to a specialist for this," no fabrication.

05 — Knowledge Base × IVR Flow · Auto-Assembly

Your data grows into a brain — and a flow

Feeding it data is flexible: paste one URL to deep-crawl the whole site, or upload large batches of mixed-format documents (PDF, Word, Excel, PPT, images, plain text). The system adapts to the content — deciding whether to split into multiple knowledge bases, auto-running scenario classification, then automatically assembling the whole conversation into a runnable flow node graph: Router, KB lookup, API integration, lead capture, transfer, end.

  • Deep website crawler — paste one URL to grab the whole site; content is auto-chunked, vectorized, and ingested.
  • Mixed-format batches — PDF / Word / Excel / PPT / images / plain text all read in.
  • Adaptive scenario classification — as content grows it auto-splits into multiple scenario KBs (product / warranty / returns) for more accurate lookups.
  • Auto-assembled flow — Router, KB, API, Lead Capture, Transfer, and End nodes are wired automatically into one flowchart.
INGEST → KB → FLOW
URL · Deep crawlPDF · WordExcel · Images
Adaptive
Scenario KB · ProductScenario KB · WarrantyScenario KB · Returns

Auto-assembled into an IVR Flow

Inbound
RouterIntent routing
KB nodeAuto-split by scenario
ProductWarrantyReturns
API nodeCheck order / member
Lead capture nodeLeave contact details
Transfer nodeTransfer to human extension
End nodeWrap-up · Hang up
KB nodeLarge content is auto-split into multiple scenario KBs; whatever is asked queries the right one.FlowRouter → KB / API / Lead Capture / Transfer → End, all nodes wired automatically.

06 — Deep Website Crawler · How It Works

Even JavaScript SPA sites — fully captured

Many sites don't hard-code their content in HTML — it's "rendered" in the browser by JavaScript (SPAs like React or Vue). Ordinary crawlers grab only the raw HTML and see an empty shell. Qubby uses a headless browser that, like a real person opening a page, runs the JavaScript and waits for dynamic content to render before capturing it — so no matter how the site is built, the content comes through.

  • Headless browser rendering — runs JS and waits for AJAX and dynamic content to load before reading, missing no SPA content.
  • Deep site exploration — follows links from the homepage layer by layer across the whole site, auto-deduplicating and staying within the same domain.
  • Noise removal — automatically filters out nav bars, footers, and ads, keeping only the main text.
  • Any format — static HTML, SPAs, infinite scroll, and multilingual pages all handled.
HEADLESS RENDER
https://spa-website.com
Raw HTML:<div id="root"></div> ← empty shell, content not yet rendered
Running JavaScript, waiting for dynamic content…
Full content captured
Static HTMLSPA · React/VueInfinite scrollMultilingual pages
RenderHeadless browser runs the JS before reading → you get the full page a real person sees.IngestWhole site crawled → noise removed → chunked and vectorized → merged into the knowledge base.

07 — AI Design Engine · Built by Conversation

No tools to learn — just talk to change your agent

Behind every service role is an AI conversational reasoning engine helping you design it. Whatever issue you hit or feature you want to add, just say it in plain language — "the greeting is too formal," "add a question about budget," "read in this new PDF too" — and the engine reasons out what you mean and adjusts the settings automatically: adding nodes, editing prompts, importing files, tuning tone. You never touch the flowchart.

  • Just talk — no need to understand flowcharts or write prompts; state what you need and it gets it.
  • Auto-applied — add nodes, change tone, import data, reasoned out in one pass and live the moment it's done.
  • Test as you go — after a change, place a test call right in the browser to confirm; if it's off, say one more line.
  • Traceable — every adjustment is logged, so you know exactly what it changed.
AI BUILDER
AI Design Assistant● Auto-adjusts to your needs
Customers often ask about business hours, please add it
Reasoning… understanding the need → knowledge base
Knowledge base updated+ Business-hours Q&A
Make the answers a bit shorter
Tone adjusted~ Style: concise
You sayState the issue or the feature you want in plain language.It doesThe reasoning engine adjusts settings automatically (nodes / tone / files), live the moment it's done.

08 — AI Reasoning · Process Docs Auto-Turned into APIs

Hand it your API docs — it learns to query your systems

That same reasoning engine can also read your systems. Hand it your API docs (or a process spec, or OpenAPI) along with credentials, and the AI automatically reasons them into individual callable tools — which endpoint, what parameters, how to authenticate. Mid-call, when a customer asks "where's my order," the AI calls your CRM / order system in real time to find the answer and speak it back. MCP servers are supported too.

  • No coding — upload the docs and it auto-groups endpoints and detects the auth method, no engineer to wire it up.
  • Real-time real data — check orders, members, inventory, appointments; it speaks the actual data in the system right now.
  • Auto member recognition — matches by phone / member ID, so on incoming calls it knows who it is and can recite their details.
DOC → TOOLS → LIVE CALL
Your API docs
Check order GET /ordersCheck member GET /memberCreate ticket POST /ticket🔑 Bearer auth
▸ Mid-call Customer: "Has my order arrived yet?"
→ call getOrder(phone:0912…)
← 200 {status:"Shipped", eta:"Tomorrow"}
▸ AI replies "It's shipped, expected tomorrow!"
ParseAI reads the docs into tool cards: endpoint + parameters + auth, fully automatic.CallReal-time call mid-conversation → gets the actual data → speaks it back in plain words.

09 — Lead Capture · Conversational Data Collection

When it's time to collect data, it asks on its own

Once you set up "which scenario, which fields," the AI asks the customer for the data in order, naturally within the conversation — not a cold form, but one question at a time. Scenarios can be anything: appointments, messages, quote requests, event sign-ups… each collecting different fields. The phone number is pulled automatically from the call itself, so the customer never has to give it.

  • Scenario-triggered — the moment a customer signals intent (wants to book / interested), it launches the matching lead capture.
  • Asks in order — one item at a time, filling whatever's missing, so the conversation stays natural, not stiff.
  • Multi-scenario, multi-form — appointments, messages, quotes each get their own fields; with a date field it can even check a calendar.
  • Auto-filing — phone auto-filled, and the list exports to Excel in one click.
SCENARIO → COLLECT
AppointmentMessageRequest quoteEvent sign-up
Lead Capture Assistant● Scenario: appointment
Detected "wants to book" → collecting in order
Of course, I'll arrange that for you — may I have your name?
Ms. Wang
And a convenient time?
Friday afternoon this week
Saved Name Ms. Wang · Phone 0912… (auto) · Time Fri afternoon
TriggerDifferent scenarios launch different forms; the AI knows which fields to collect right now.CollectAsk one by one until complete → phone auto-filled → written to the list, exportable.

10 — AI Auto-Survey · How It Works

It decides the next question based on the answer

Drop in your survey document and the AI automatically detects question types (single-choice / open-ended / rating) and options, laying them out into a branchable flowchart. Once the outbound call connects it asks question by question, deciding which question to jump to based on the answer — if they're satisfied it asks whether they'd recommend, if not it probes the reason — then compiles everything into a summary and files it.

  • Conditional skips — in one survey, different people take different paths and aren't asked irrelevant questions.
  • Randomized question sets — question order can be shuffled to reduce order bias.
  • Open-ended understood too — not just button-presses; respondents speak freely and the AI distills the key points.
  • Exportable results — per-response answers + summaries, downloadable as CSV / reports.
BRANCHING FLOW
Opening intro
INTRO
Q1 Satisfaction?
Rating 1–5
Q2 What's wrong
≤ 3
Q2 Recommend?
≥ 4
End · Save summary
SUMMARY
ExampleRespondent answers "5" → system auto-takes the right branch, asks "would you recommend us," skips the complaint question.

11 — Batch Outbound Campaigns · How It Works

One list, the scheduler dials every call automatically

Import a list, set the calling hours, and the scheduler runs automatically. How many calls at once = how many extensions you check (one call per extension); every call detects whether it truly connected or hit voicemail and reports the classified result in real time. Calls that don't get through are auto-redialed by the rules you set.

  • Only true connects count — detects ringing / SIP signaling to judge pickup, never treating voicemail as a person.
  • Redial control — ring duration, redial interval, max attempts, and numbers per person are all configurable.
  • Variable-driven greetings — {{name}}{{title}} pulled from the list, and different greetings can play by category.
  • API-triggered — external systems can call a single outbound (appointment reminders, notifications), triggered in real time.
SCHEDULER
ListMs. Wang 0912…Mr. Chen 0928…Manager Lin 0937…+ 1,240 entries
Concurrent lines (= number of extensions checked)
Extension 801
Connected ✓
Extension 802
No answer · Redial
Extension 803
Voicemail
Concurrency3 extensions → 3 calls at once; want it faster, open more extensions.ResultsConnected / no answer / busy / voicemail classified in real time; no-answers enter the redial queue.

12 — Outbound Personalization · CRM × Dynamic Greetings

Everyone hears the line tailored just for them

Lists can be imported straight from your CRM (or triggered one by one in real time via API). Write the greeting as a template with variables like {{name}}{{title}}, and each person's details are filled in automatically at dial time. Going further: based on the category in the list, one campaign can run different greetings, different flows, and collect different fields for different people — VIPs hear priority service, regular customers get a satisfaction survey.

  • CRM integration & import — lists come in from your CRM / Excel, or trigger a single outbound in real time via API.
  • {{variable}} dynamic greetings — {{name}}, {{title}} and more filled in automatically; missing values are left blank, no errors.
  • Category branching — by the category field, pick different greetings / take different flows / collect different fields.
  • One campaign, many scenarios — cold-snap check-ins, typhoon notices, renewal reminders… one campaign handled separately by category.
CRM → {{ }} → Category branching
CRM / Excel listFields: name · title · category · topic
Greeting template "Hello {{name}} {{title}}, this is OO Company, regarding your {{topic}}…"
↓ Fill from list data + route by category
Manager Wang Da-ming VIP
"Hello Manager Wang Da-ming… regarding your contract renewal, we have a specialist to serve you with priority"
→ VIP flow · Collect: renewal intent
Chen Xiao-fang Regular
"Hello Chen Xiao-fang… we'd like to run a short satisfaction survey with you"
→ Regular flow · Collect: satisfaction rating
Variables{{name}}{{title}} pulled from the list → each person hears their own name and context.BranchingOne campaign by category → different greetings, different flows, different fields collected.

13 — Web Service · One-Line Embed

The same AI, lives on your website too

Not just the phone. Paste one line of <script> into your site and a service widget floats up in the bottom-right; or just share a public link. Visitors can chat with it by voice or text, backed by the same role and the same knowledge base. When it needs a human, the web conversation transfers straight to a specialist's extension on your PBX.

  • One-line embed — floating bubble or inline; login and mic permission are handled by us, no allowlisting and no backend changes for you.
  • Voice + text dual mode — speak or type, the same AI and the same data.
  • Seamless handoff to a human — web voice transfers to a specialist using the same bridging / REFER mechanism as the phone.
  • Your brand colors — public links, widgets, and share pages unified visually.
WEB WIDGET
<script src="https://ivr-web.qubby.io/embed.js" data-agent="your-id"></script>
https://yourcompany.com
Live Support● Voice / Text
VoiceText
I'd like to ask about installationSure, let me check your area~I'd like to talk to an agentTransferred to agent · ext. 302
EmbedOne line of script → floating widget or inline; Firebase all on our origin, no allowlisting on your side.To a humanWeb voice → your PBX extension, using the same transfer mechanism as incoming calls.

14 — Human-Agent Notifications · Chrome Extension

Hand off to a specialist — with the full context

The human-agent Chrome extension is now live. Specialists log in with their account and bind their own extension. When the AI transfers a call to that extension, the extension pops an instant notification, pushing the customer name, phone, AI conversation summary, and transfer reason to the specialist — so the moment they pick up they know what the customer just discussed with the AI and why it's being transferred, no need to start over.

  • Extension binding — specialist logs in with account + extension (ext); when the AI transfers to that extension, the notification arrives.
  • Full context delivered — customer name, phone, AI conversation summary, and transfer reason, all at once.
  • Seamless takeover — the specialist needn't re-ask, just continues from where the AI left off.
Now live · Chrome Web Store
FLOW + AGENT POPUP
AI on call
Transfer ext 302
Push info
Agent Chrome
Human-agent notificationLogged in · Extension 302
A call is being transferred to you
CustomerMs. Wang
Phone0912-345-678
AI conversation summaryAsked about installation service, confirmed the warranty period, and wants to schedule an on-site time directly with a specialist.
Transfer reason: customer requested a human
AnswerView full text
FlowAI transfers to extension 302 → simultaneously pushes the customer's data and summary to the specialist bound to that extension.InterfaceChrome extension pops a notification: name / phone / AI summary / transfer reason, answer in one tap.

15 — WebCall · Multi-Entry Voice (WebRTC)

Web, Line, or App — reach the AI by voice in one tap

Beyond embedding on your site, Qubby also lets users call the AI directly by voice — from the web, Line (LIFF), or your App, one tap connects over WebRTC in 16kHz wideband HD audio (clearer than a regular phone). When the customer needs a human, the AI transfers this web call to a real specialist's extension on your PBX, just as seamlessly as a phone call.

  • Multiple entries — web, Line LIFF, and App can all start a voice call; users need no phone and no dialing.
  • Direct WebRTC — native real-time voice in the browser / App, no plugin, no download.
  • 16kHz wideband HD audio — far clearer than an 8kHz PSTN phone, sounding more natural.
  • Transfer to an extension — when needed, route the web call to a real internal PBX extension, integrated with your existing PBX.
WEB · LINE · APP → AI → PBX
WebBrowser
LineLIFF
AppSDK
WebRTC voice · 16kHz wideband
Qubby AI
Answer · Converse · Look up data
When a human is needed · Transfer
PBX human extension
Internal extension · Integrated with PBX
♪ 16kHz HD VOICE
EntryFrom web / Line LIFF / App, tap once to place a voice call to the AI over WebRTC.TransferCustomer wants a human → this web call routes to a real internal extension on your PBX.

16 — Call Records · Transcripts · AI Summaries

Every call — seen, heard, and understood

Every call (incoming, outbound, web) automatically leaves a recording and transcribes the conversation to text. If a transfer to a human goes through ConfBridge, even the specialist-and-customer conversation is recorded and transcribed. The AI then generates a summary and intent analysis for each call and can apply tags to classify them — total usage and outcomes at a glance.

  • Full recording + transcript — every call replayable (1–2x), speech-to-text with a correctable transcript.
  • Human conversation recorded too — after a ConfBridge transfer, the specialist ↔ customer conversation is recorded and transcribed by segment.
  • AI summary + intent — auto-generated summary, with intent / outcome / sentiment judged.
  • Tags + dashboard — custom tags for classification; KPIs, usage, and end reasons at a glance.
RECORDING · STT · SUMMARY
Incoming0912-345-678Sales list03:42
Cust.How do I book your installation service?
AIYou can book online, let me check your area~
After transfer · Human ↔ customer (ConfBridge recording)
AgentHello, I've scheduled you for Friday afternoon at your place
Cust.Great, thanks
AI summary: Customer asked about installation service and requested a specialist; transferred, scheduled for Friday afternoon at their place.
Intent: book installationSentiment: positiveOutcome: transferred to human
TranscribeThe AI segment and (post-transfer) human segment are both speech-to-text; transcripts can be replayed and compared.AnalyzeEach call auto-generates a summary, judges intent / sentiment / outcome, and can be tagged into the dashboard stats.

Plans & Pricing

Two plans, priced by "seat"

Pick one based on whether you want a pure transfer receptionist or a service agent that looks up data and captures leads. Pricing is by seat (how many concurrent lines) — open as many as you use.

AI Smart Receptionist
RECEPTIONIST · Pure Transfer
NT$499/ line / month
No setup fee · Unlimited call minutes
  • Answer, ask who they want, look up the directory and transfer
  • Same-name / not-found fallback handling
  • Full recording · transcript · AI summary
  • KB Q&A / lead capture / API integration
  • Outbound campaigns / surveys
Most complete
AI Smart Service
CUSTOMER SERVICE · Full-Featured
NT$6,000/ seat / month
Setup fee NT$30,000 (one-time) · Includes 800 min/seat, NT$3/min overage
  • Includes all AI Receptionist features
  • KB Q&A (RAG, grounded answers)
  • In-call lead capture · list export
  • API / MCP integration (orders, members…)
  • Batch outbound campaigns · automated surveys

Overseas converted by local currency (USD basis: receptionist seat $15, service seat $200, overage $0.1/min) · Actual plan per quote

Want to hear how it answers your calls?

Give us 30 minutes and a specialist will walk you through your real scenario — watch the AI answer the call, look up data, and transfer to a human, end to end.

Book a consultation