Skip to content
Need A Website? Get One Up With No Money UpfrontLearn More
Cylas Tee

AI Chatbot & Assistants

An Assistant That Reads Your Documents Before It Answers

Can you survive without a chatbot? Yes, you have so far. But your customers ask the same twenty questions, at every hour, in every time zone, and the ones who ask at midnight have found another answer by the time you wake up. We build an AI assistant that answers from your own pages, FAQs and PDFs, shows where the answer came from, captures the lead and hands over to a real person the moment it is out of its depth. On your website, on WhatsApp, on Telegram, running on Claude or OpenAI through an account you own.

Build My Chatbot

Usually replies within 12 hours · Fixed quote before starting

Send Us A Message

Tell Us About Your Project

  • More Than A DecadeIn The Digital Space
  • 9Countries Served
  • 12hUsual Reply Time
  • 100%Remote-First

The Unfair Advantage

The Difference Between A Chatbot And A Chatbot That Knows Your Business

What if I tell you the problem with most AI chatbots is not that they are dumb? It is that they are confident and have never read a word about your business.

A general AI model knows a lot about the world and nothing about your return policy, your pricing tiers or the way you explain your programme. Ask it anyway and it will guess, fluently, which is worse than saying nothing. The fix has a technical name, RAG, and a plain meaning: the bot reads your documents before it answers. Every question first pulls the most relevant passages from your own content, and only then does the model write a reply, in your words, with the source attached.

That one change turns a novelty into a staff member. It answers what you have already written down, it says “let me get someone” for what you have not, and it never promises a refund you did not approve. We build that for you, train it on content you sign off, and leave it running in accounts in your name.

Build My Chatbot

Advantage #1

It Answers In Your Words, From Your Content

We start with what you already have: your website pages, your FAQ, the PDFs you send every new client, the email replies you have typed a hundred times. That content gets broken into passages and stored in a vector database, which is simply a library that can find passages by meaning rather than by exact word. Ask “do you do payment plans?” and it finds the paragraph about instalments even if the word “plan” never appears.

When a customer asks something, the assistant looks up the handful of passages most related to the question, hands them to the model with strict instructions to answer only from them, and replies. Below the answer it can show which page or document it used, so you, and your customer, can check it.

Did you know? The same approach works inside your business too. An internal assistant trained on your procedures, templates and past projects answers your team’s questions the same way, and nobody has to interrupt you to find the onboarding checklist.

Advantage #2

It Knows When To Stop And Fetch A Human

The best thing an assistant can say is sometimes “I am not sure, let me get Cylas.” We set guardrails before launch: the topics it must never answer on its own, such as refunds, legal, medical or pricing exceptions, the moment it should hand the conversation to a person, and what it does if someone tries to talk it into something silly.

Handover is not a dead end. The customer gets told a human will follow up, you get the full conversation in your inbox, Slack or CRM with the lead’s details already captured, and the thread continues where it left off. The assistant did the triage. You do the part that needs you.

Advantage #3

It Lives Where Your Customers Already Are

A chat bubble on your website is the obvious start, and for most businesses the right one. But if your customers message you on WhatsApp or Telegram, the assistant can answer there too, from the same content, with the same rules. One brain, several doors.

Every conversation is logged. After launch we read the real ones with you, see which questions it handled well, which it should have handed over sooner and which your content does not answer yet, then update the documents and the guardrails. It is a win-win scenario: the assistant gets better every month and your FAQ page does too.

Who It's For

Who This Is For

If your inbox has the same questions in it every week, an assistant trained on your answers is the obvious next step.

  • Coaches & Course Creators

    Programme details, payment options, “is this right for me?” The assistant answers from your own sales page and FAQ, then books the call when the lead is warm.

  • E-Commerce Stores

    Shipping, sizing, returns and “where is my order?” at any hour, from your own policies, with the tricky ones handed to a person.

  • Service Businesses

    What you do, where you do it and what it roughly costs, answered instantly on your site or WhatsApp, with the enquiry captured into your CRM.

  • Teams With A Pile Of Documents

    An internal assistant trained on your procedures, templates and past projects, so your team stops asking you where things are.

What's Included

A Complete Assistant, Trained And Supervised

Everything from the first document to the first real conversation review.

  • Content Audit & Training Set

    We gather your pages, FAQs, PDFs and standard replies, fill the gaps you did not know you had and get your sign-off on exactly what the assistant may use.

  • Vector Database & RAG Pipeline

    Your approved content indexed by meaning in Supabase pgvector, Pinecone or Qdrant, so every answer starts by reading the right passages.

  • Assistant Personality & Guardrails

    Tone, boundaries, the topics it must never answer alone and the exact moment it hands over. Tested on your awkward questions before launch.

  • Website Widget & Messaging Channels

    A chat widget on your site, and WhatsApp or Telegram if your customers live there, all answering from the same content.

  • Human Handover & Lead Capture

    Name, contact and the full conversation delivered to your inbox, Slack or CRM, with a Telegram alert for hot leads.

  • Conversation Review & Tuning

    We read the real conversations with you after launch, fix what it got wrong and add what your content was missing.

Our Chatbot Build Process

From A Pile Of Documents To An Assistant You Trust

Short sprints, recorded demos and your own awkward questions as the test set. Fully remote, calls in your time zone.

  1. 01

    Questions & Content

    We start with the twenty questions you answer most, then look at what you already have written down: pages, FAQs, PDFs, the email you keep forwarding. Usually a few answers exist only in your head, so we write those down too. You approve the set, which becomes the only thing the assistant is allowed to answer from. We also agree the handover rules here, because deciding what the bot should not do is more important than deciding what it should. You get a fixed quote before anything is built.

  2. 02

    Design The Conversation

    Before any code, you see a plain-English diagram of how a question becomes an answer: where it looks, what it is told, when it stops and fetches a person, and where the lead ends up. We write the assistant’s tone and boundaries with you, in your voice, so it says “we” the way you do. Changing a rule on paper takes minutes. Changing it after a customer has screenshotted the wrong answer takes a lot longer.

  3. 03

    Build, Then Try To Break It

    Your content goes into the vector database, the assistant goes onto a test page, and we hand you the keys with one instruction: ask it the questions you are worried about. Pricing exceptions, refunds, the competitor comparison, the message that tries to trick it. At the end of each sprint you watch a recorded demo of the real thing answering real questions with its sources shown. We tune until you would let it talk to a client.

  4. 04

    Launch & Read The Conversations

    The widget goes on your site, the messaging channels are connected, lead capture and alerts are switched on and the walkthrough is recorded so you can update a document yourself. Then we let it run for a few weeks and read the real conversations together, because real customers phrase things in ways no test set predicts. You keep it running yourself, or keep us on and we keep reading, tuning and expanding what it knows, quoted monthly.

AI Chatbot & Assistants Pricing

Scoped With You, Quoted Once

No two assistants are the same, because no two businesses have the same content, channels or rules. So we scope it with you first and give you a fixed quote before anything is built.

You can switch on a generic AI widget in an afternoon. What you cannot do in an afternoon is gather and approve the content it answers from, set the rules for what it must never say, wire the handover to a person and read the first month of real conversations.

That is the part that makes it safe to put in front of paying customers, and it is the part we quote for, once we have seen your content and agreed the channels. The way in is a short call or the free website audit. Tell us which questions your customers ask most, and you will have a fixed price in writing before anything starts.

Website Assistant

An assistant trained on your approved content, with guardrails, lead capture and human handover, as a chat widget on your website.

From$1,997USD one-time

  • Content audit & approved training set
  • RAG with a vector database you own
  • Website chat widget
  • Guardrails & human handover
  • Lead capture to your CRM with hot-lead alerts
  • Conversation review & tuning after launch
Build My AI Chatbot

Prices are starting points. Assistants are scoped first, then quoted as a fixed price. AI model usage (Claude or OpenAI), vector database and messaging API fees are paid directly by you. Builds: 50% deposit, 50% on launch.

FAQ

Frequently Asked Questions

Straight answers to the questions we hear most. If yours is not here, just ask.

What Does RAG Actually Mean?
RAG stands for retrieval-augmented generation, which is a long way of saying the bot reads your documents before answering. When a question comes in, the assistant first retrieves the most relevant passages from your own content, then writes its reply using only those passages, with the source attached. Without RAG, a chatbot answers from whatever the general model happens to know, which is nothing about your business.
What Is A Vector Database, And Why Do I Need One?
It is a searchable library of your content that finds passages by meaning rather than by exact word. Ask “can I pay monthly?” and it finds your paragraph about instalments even though the word “monthly” never appears. We use Supabase pgvector, Pinecone or Qdrant depending on your setup, in an account you own. It is what makes the assistant fast, accurate and easy to update: change the document and the answers change with it.
What Can The Assistant Be Trained On?
Anything you approve: website pages, FAQs, PDFs, brochures, course outlines, policy documents, the standard replies you send by email. We also help you write down the answers that only exist in your head. Nothing goes in without your sign-off, and nothing is shared with other clients.
Will It Give My Customers Wrong Answers?
Any AI can, which is why the build is mostly about preventing it. The assistant answers only from your approved content, shows its sources, is told which topics to never answer alone, and hands over to a person when it is unsure. After launch we read the real conversations with you and tighten anything that slipped. Human handover is part of every build, not an extra.
Which Channels Can It Run On?
A chat widget on your website is the usual starting point. It can also answer on WhatsApp and Telegram through their business APIs, and we can build an internal version for your team. The same content and the same rules apply on every channel, so you train it once.
Claude Or OpenAI? Which Model Do You Use?
Either, through an API account in your name. We recommend one based on how your content reads and how carefully the assistant needs to behave, and we will tell you why. Because the content lives in your own vector database, switching providers later is a configuration change, not a rebuild.
Is My Data Private?
Your content and your conversations stay in accounts you own: your model API key, your vector database, your messaging accounts. We do not train anything on your data for other clients, and you can revoke our access or switch the assistant off at any time without anything else breaking.
How Long Does It Take, And What Does It Cost To Run?
A website assistant typically goes live in 2–4 weeks, including gathering the content and testing. Adding WhatsApp or Telegram typically adds a week, mostly waiting on the business API approvals. After launch you pay the model usage, the vector database and any messaging fees directly to the providers. We estimate those up front, and for most small businesses they are modest.

Our Guarantees

Promises We Put In Writing

We would rather work for free than see an unhappy client. These are the guarantees every project comes with.

Fixed Quote Before Work Starts

Automation work has a habit of growing. One workflow becomes three, and the invoice grows with it. So we do it differently. After the process audit you get a fixed scope and a fixed price, in writing, before anything is built. If you want more later, we quote it before we touch it, not after. The only ongoing costs are the software and AI usage fees you pay directly to the providers, and we estimate those up front too. No surprises, no hidden extras.

You Own The Code And The Accounts

Everything is built inside accounts in your name: your Claude or OpenAI API key, your vector database, your WhatsApp or Telegram business account and the website the widget sits on. Nothing runs through a login you cannot see. If you want, just send us “We will take it from here”, we won’t take it personally, and everything keeps running exactly as it did, because none of it ever depended on us. We would rather you understand your own systems than be stuck with us.

Our Time Is Limited

Mapping a business properly takes time and attention, and Cylas Tee or someone from his team reads and tests every workflow before it goes near your customers. To make sure each client gets that attention, we only take on a handful of new projects each month. If you are remotely interested in getting your admin off your plate this quarter, we strongly suggest you contact us today to reserve a slot.
Reserve Your Slot

Got A Question?

Ready To Put Your Answers To Work?

Tell us which questions your customers ask most and where they ask them. You will get a plain-English plan, a recommended setup and a fixed quote, and we will tell you honestly if a good FAQ page would do the job instead. Every message is read and answered by Cylas Tee himself or someone from his team.

Build My Chatbot

Usually replies within 12 hours · Fixed quote before starting