Skip to content
1

The chat that  knows what your business already knows

The problem was never the chat interface — it was what the chat was grounded in. Vector Stores indexes your complete document corpus into a semantic knowledge layer that Live Chat queries at runtime, by meaning. The chat sounds like you because it draws from what you've actually documented.

Semantic matching, not keyword Full corpus indexed Zero hallucination grounding
Live Chat
What's included in your premium package?
Typical chat
I don't have that information. Would you like to schedule a call with our team?
Vector Stores
The Premium package includes unlimited consultations, quarterly portfolio reviews, priority scheduling, and a dedicated advisor. Annual billing includes two complimentary estate-planning sessions.
Semantic match
Source: premium-service-guide.pdf — §3.2
02

Your business knows the answer. Your chat doesn't.

The knowledge exists. It's documented. It's sitting in a folder somewhere — while your chat operates without it.

Fiduciary Advisor
"What does your fee actually cover?"

You have a detailed service model document on your drive. Your prospect is asking at 10pm on a Sunday. Your chat has no idea. They leave. You never know they were there.

Mortgage Broker
"Which banks work with contractors?"

Your lender comparison sheets cover every product. A self-employed visitor needs one specific answer. The chat returns nothing useful — so they try the broker whose site actually helped.

Coach / Consultant
"How long is the full engagement?"

Your programme structure document has every detail. The chat guesses or deflects. The prospect wanted an answer now — not a calendar link for next Tuesday.

Private Practice Lawyer
"Do you handle messy divorces?"

Your practice area descriptions are written in precise professional language. The visitor asked in their language. The chat cannot bridge that gap.

Product Details
Pricing & Fee Structures
Policies & Compliance
FAQs & Process Guides
No retrieval path
What does your fee actually cover?
I'm sorry, I don't have that information. Please schedule a call →

Every "I don't know" is a prospect who leaves your site. Every hallucinated answer is a trust violation you may never learn about. Every deflection to "schedule a call" filters out the visitors who wanted an answer now — not next Tuesday. You're paying for website traffic and then letting an uninformed chat squander it.

3

FAQ databases don't scale. Generic AI doesn't know you. There's a third option.

Two approaches dominate. Both are structurally inadequate — not because the people using them are wrong, but because the architecture underneath is incapable of the job.

The FAQ Database

Works for ten questions. Breaks at fifty. Becomes unmaintainable at a hundred. And it fails completely when a visitor phrases their question differently than the FAQ expected — which is most of the time. Keyword matching is brittle by design. The more you build, the more it fragments.

Brittle & keyword-dependent

The Generic AI

Sounds fluent. Sounds confident. Has no knowledge of your specific policies, pricing, services, or processes. When it doesn't know, it fills the gap with plausible-sounding fabrication. You can't verify what it told a visitor at 2am — and the visitor can't know the answer was invented.

Fluent but ungrounded
The third option

Vector Stores + Live Chat

A knowledge layer that indexes your real documents and lets Live Chat retrieve answers by meaning — grounded, current, and scoped to the right context. Not a bigger FAQ. Not a smarter chatbot. A different architecture that connects what visitors ask to what your business actually knows.

Grounded in your knowledge
FAQ Database
? No keyword match
Generic AI
No document source
Vector Stores + Live Chat
"What's your process?"
Grounded
4

Ground the chat in what you actually know. Not in what AI thinks you might say.

The chat's quality ceiling is set by the knowledge it has access to. Give it nothing, and it guesses. Give it a small FAQ, and it handles the easy questions. Give it your complete indexed knowledge — every document, policy, product sheet, pricing table, process note — and it can answer the way you would.

Because it draws from the same source material you do.

Index once, retrieve continuously. You don't teach the chat responses. You index your documents — in whatever format they already exist — and the chat retrieves from them by meaning at the moment a visitor asks.
Knows its own boundaries. If the knowledge exists in an indexed document, the chat finds it. If the knowledge hasn't been indexed, the chat knows the boundary of what it can say.
Your role is your knowledge. Maintain your documents. The system's role is to make that knowledge findable — semantically, at runtime, for every visitor who needs it.
5

Your chat finds answers by what they mean, not what they say

Keyword search requires visitors to guess the exact words your documents use. If they say "premium plan" and your document says "Tier 3 package," keyword search returns nothing. Semantic search understands that these phrases mean the same thing — and delivers the right answer regardless of how the question is phrased.

"What's included in the premium plan?"
Keyword Search
Searches for exact match: "premium plan"
Scans documents for those literal words
No document contains "premium plan" — only "Tier 3 package"
No results found
Semantic Search
Decomposes question into meaning: highest-tier offering + contents
Matches meaning against indexed knowledge — not words
Finds "Tier 3 package" as conceptual equivalent
Grounded answer delivered
meaning match: 94.7%

Every visitor phrases it differently. Semantic search handles all of them.

Your business knowledge is not too complex for AI — it just needs to be indexed by meaning, not by keyword. Here's how it works across four real scenarios.

Mortgage Broker
Visitor "Which lender has the lowest rate for a rental property?"
Document "Lender B — Investment Property Program: 6.75% for non-owner-occupied residential"
Bridge "rental property""non-owner-occupied residential" — same concept, different vocabulary

The visitor doesn't know your lender's terminology. Semantic search does. The comparison sheet is indexed once — every phrasing resolves.

Coach / Consultant
Visitor "Do you offer one-on-one sessions or is it all group stuff?"
Document "Accelerator Track: includes 6 private coaching sessions + community cohort access"
Bridge "one-on-one sessions""private coaching sessions" — the visitor's casual phrasing meets your programme structure

No FAQ could anticipate "group stuff." But the meaning — individual vs. collective formats — is clear to semantic retrieval.

Private Practice Lawyer
Visitor "Can you help if my business partner is trying to push me out?"
Document "Practice Areas: Shareholder disputes, minority oppression claims, partnership dissolution"
Bridge "push me out""minority oppression claims" — emotional language maps to legal terminology

Visitors describe problems in plain language. Your practice area pages use legal terms. Semantic search bridges the gap without you maintaining a glossary.

Fiduciary Advisor
Visitor "Do you manage money for people who are about to retire?"
Document "Service Model: Pre-retirement income transition planning for clients within 5 years of target retirement date"
Bridge "manage money" + "about to retire""pre-retirement income transition planning" — casual intent meets fiduciary precision

Your service model is precise because it has to be. Your visitor's question is casual because they're exploring. Semantic search respects both.

6

Multiple stores, one chat: the right depth for the right audience

You would not hand a casual website visitor your internal operations manual. You should not hand your client-facing chat access to everything in one undifferentiated pool. Vector Stores lets you organise knowledge into distinct contexts — and Live Chat draws from the appropriate store based on who is asking and what they need.

Customer-Facing Store

Service descriptions, pricing, common questions — the default for general visitors. Clear, helpful, appropriate.

Product-Specific Store

Deeper detail about a particular offering — coverage matrices, technical specifications, detailed comparisons. Referenced when the visitor is clearly asking about a specific service.

Internal Store

Operational knowledge not appropriate for client-facing channels — carrier pricing logic, internal workflows, margin structures. Indexed for completeness but never surfaced to visitors.

Common assumption

"One knowledge base is enough." It is — until your chat tells a prospect your internal carrier pricing, or gives a general visitor the same technical depth as a qualified lead. Precision is not complexity. It is control over what reaches which channel.

Scenario: Insurance principal
07

Update a document. Every chat response reflects it instantly.

When you change a document — a policy update, a price adjustment, a new service description — Live Chat retrieves from the updated content at runtime. The next visitor query reflects the change. No cache to clear. No chat to retrain. No configuration to touch.

This is the architecture resolving a quiet anxiety: in fast-moving industries — mortgage brokers with rate changes, advisors with fee adjustments, accountants with regulatory updates — your chat is never operating on yesterday's information. Runtime retrieval is that confidence.

Real estate scenario

A brokerage policy changes Monday night. The realtor updates their commission structure document Tuesday morning. By Tuesday afternoon, every Live Chat response touching on commission structure reflects the new information — without the realtor doing anything beyond updating the document itself.

Common assumption

"I'll just update my AI's responses manually when something changes."

One document might inform dozens of chat responses. Runtime retrieval means the update propagates to every response that touches that knowledge — automatically, immediately, and without you mapping which responses depend on which facts.

Commission Policy 9:14 AM
Live Chat
What's your commission structure?
Our standard commission is 2.5% with flexible arrangements for repeat clients.
retrieved at runtime
8

Upload what you have. In whatever format it exists.

No reformatting. No restructuring. No technical configuration. Upload the documents your business already runs on — and Vector Stores indexes them instantly. Every document makes Live Chat smarter.

PDFs
Spreadsheets
Text docs
Product sheets
Pricing tables
Policy manuals
FAQs
Process docs
Training material
Knowledge compounds

Five documents on day one. Two more per month. A transformed chat by quarter's end.

Every document you index makes Live Chat more capable — without reconfiguring anything. Start small. The knowledge compounds on its own.

5 Day 1
7 Month 1
11 Month 3
17 Month 6
Any format in → Grounded answers out
Pricing Guide Indexed
Service FAQ Indexed
Coverage Matrix Indexed
Onboarding Guide Indexed
Vector Store
Live Chat response
"Our premium tier covers all three regions — here's the pricing breakdown for your team size."
9

How Vector Stores works alongside Reusable Knowledge

Curated items are your best thinking in your exact words. Indexed documents are your comprehensive corpus, retrievable by meaning. Live Chat uses both — curated first for precision, Vector Stores second for breadth.

"Do you handle properties in the Blue Mountains?"
Reusable Knowledge
Vector Stores
Yes — I've handled 12 transactions in the Blue Mountains region over the past 18 months, including…
↑ Sourced from: Vector Stores — Service Area Guide.pdf
Reusable Knowledge

Your curated content library — objection responses, talking points, FAQs, rapport questions, conversation starters — hand-crafted or approved in your exact language. These are the high-priority responses that represent your best thinking on the topics that matter most.

Checked first · Precision match · Your exact words
Vector Stores

The semantic retrieval layer over your full document corpus. When a visitor asks something whose answer lives in an indexed document rather than a curated item, Live Chat queries Vector Stores for the closest semantic match and grounds its response in that retrieved passage.

Checked second · Semantic retrieval · Full corpus breadth
Inside Live Chat, they work together

When a visitor asks a question, Live Chat checks Reusable Knowledge first. If a curated item matches, that's the response — because it's your approved language for high-priority topics. If no curated item applies, Live Chat queries Vector Stores for the closest semantic match in your indexed corpus. Curated first for precision. Vector Stores second for breadth and edge cases.

10

One source of truth. Every channel consistent.

When three visitors ask your Live Chat about pricing, service inclusions, and process timelines within the same hour, every response draws from the same verified knowledge base. There is no scenario where one visitor hears one thing and another hears something different — because the source material is the same indexed corpus.

In relationship-driven industries, inconsistency is a trust killer. If a prospect's chat interaction contradicts what they later hear from you directly, the damage is disproportionate to the error. Vector Stores eliminates this structurally — your chat and your voice draw from the same knowledge.

Update your knowledge once. Every subsequent Live Chat interaction reflects that single source of truth — immediately, without retraining, without gaps.

Structural consistency, not policy compliance. Your reputation is protected by architecture, not by hoping the AI gets it right.
Vector Store
Visitor A
"What does your full-service package include?"
Grounded
Visitor B
"How much do you charge for an initial consultation?"
Grounded
Visitor C
"What's the typical timeline from start to close?"
Grounded
11

Common questions about Vector Stores

The opposite is true. Vector Stores use semantic indexing — meaning the system understands the meaning of your content, not just keywords. The more specialised your knowledge, the more valuable this becomes, because generic search would miss the nuance entirely. A niche technical document that would confuse a keyword search becomes precisely retrievable when indexed semantically. Specialisation is the use case, not the limitation.

No. Upload your documents in whatever format they already exist — PDFs, Word files, text documents. The semantic indexing process handles the rest. You don't need to rewrite, reformat, or restructure anything. The system reads your content as it is and makes it retrievable by meaning.

Multiple stores give you precision over what knowledge reaches which audience. A public-facing store might contain your service descriptions, pricing philosophy, and FAQs — everything a website visitor should access. A separate internal store holds operational procedures, team playbooks, or sensitive data that should never surface in a chat widget. Without this separation, you're choosing between exposing everything or withholding everything. Multiple stores let you be specific.

"Works fine" usually means the chat gives plausible-sounding answers. Without verified knowledge to draw from, it's generating responses from general training data — which means it's guessing when it doesn't have your specific information. The problem is that you won't always know when it guesses wrong. A visitor gets a confident answer about your services that's subtly inaccurate, and neither of you catches it. Vector Stores replace guessing with retrieval from your actual documents.

Setup time is proportional to how many documents you're uploading, and the process itself is closer to uploading a file than configuring a database. Create a store, drag in your documents, and the indexing happens automatically. Five documents might take a few minutes of your time. Fifty might take an afternoon. There's no schema design, no field mapping, no technical configuration required.

Semantic retrieval finds the closest match to a visitor's question and surfaces the verified content from your documents. It doesn't interpret, rewrite, or editorialize your material. The AI uses your words as the source, not as a prompt for creative generation. The retrieval step is mechanical — find the most relevant passage — and the response is grounded in what that passage actually says.

Small teams have the most to gain. When institutional knowledge lives in one or two people's heads, it's unavailable to your chat, your website visitors, and anyone who asks a question outside business hours. Even five well-indexed documents — your service overview, pricing guide, onboarding process, common objections, and a capability summary — transform what Live Chat can do. You're not building a library. You're making what you already know accessible.

No retraining required. Vector Stores use runtime retrieval, which means the system queries your documents live at the moment a question is asked. When you update or replace a document, the next query automatically uses the updated content. There's no model to retrain, no cache to clear, no deployment step. Replace the file and the knowledge is current.

A chat that finally knows what the business knows — and compounds in capability every day.

Every document indexed makes your Live Chat more capable

Your institutional knowledge — scattered across documents, drives, and team members' heads — becomes a unified, searchable foundation. Every update propagates instantly. Every new product, policy, or service description becomes immediately accessible. The chat gets better every week, not because it is retrained, but because the knowledge base grows.

Documents indexed
in real time
Retrieved
by meaning
Scoped to
the right context
Delivered
at runtime
Compounds
continuously