Why We Built JestBot: One AI Bot for Chat, WhatsApp, Calls, and Everything In Between

The problem we kept running into

Every business we talked to before building JestBot had roughly the same setup, just with different vendor logos. A website chat widget from one company, handling only website visitors. A WhatsApp Business number that a support agent answered manually from their phone, with no memory of what the customer had already asked on the website. An old IVR or calling flow, built years ago by someone who has since left the company, that nobody wants to touch. And underneath all of it, a shared spreadsheet or a wiki page of FAQs that three different tools were all trying to represent in three different ways.

None of these systems talked to each other. A lead who asked about pricing on WhatsApp on Monday had to repeat the entire conversation to a human on a call on Wednesday. A customer who got a wrong answer from the website widget had no way of knowing the WhatsApp bot (if there even was one) might know better. Every channel was its own silo, with its own training data, its own dashboard, and its own bill.

The fragmented stack anti-pattern

If you map out a typical mid-size company's customer-facing AI stack today, it usually looks like this:

• A website widget vendor, trained on a static export of the help center
• A separate WhatsApp Business API reseller, configured with its own keyword-based flows
• A call center IVR that was never meant to understand natural language
• An internal support team fielding the requests that fall through the cracks of all three

Each of these tools needs to be updated separately whenever a policy changes, a product launches, or a price moves. In practice, that means they drift out of sync within weeks. The WhatsApp bot still quotes last quarter's pricing. The website widget doesn't know about the new SKU. Nobody trusts any of the automated channels, so everything eventually gets escalated to a human anyway — which defeats the point of automating in the first place.

The real cost of disconnected channels

It's easy to underestimate how expensive this fragmentation actually is until you add it up. Every extra tool is a separate subscription, a separate login, and a separate person who has to remember to update it. Every gap between tools is a place where a lead goes cold, because nobody followed up in time, or a customer gets a wrong answer and quietly churns instead of complaining. And every inconsistency between channels chips away at trust — a customer who catches your WhatsApp bot contradicting your website doesn't conclude "the WhatsApp bot must be outdated," they conclude "I can't trust anything this company's bots tell me," and start double-checking everything with a human, which erases the entire point of automating support.

What JestBot actually is

JestBot is a single AI bot — trained on your documents, FAQs, and product catalog — that you can deploy across a website widget, a WhatsApp number, a mobile app via SDK, and an AI calling agent, all sharing the same knowledge base, the same conversation history, and the same tools. You configure the bot's personality, instructions, and knowledge once. Every channel you turn on inherits that configuration automatically.

That single point of configuration is the whole idea. When you update a document in the bot's knowledge base, the change is live everywhere at once — the widget, WhatsApp, the mobile SDK, and the calling agent — without re-training anything or redeploying five separate systems.

Built on Retrieval Augmented Generation

Instead of letting the underlying language model hallucinate answers from its general training, JestBot retrieves the most relevant chunks from your uploaded documents before generating a reply, so answers stay grounded in what you actually told it. This is the same Retrieval Augmented Generation (RAG) pattern used by most production AI systems today, and it's the difference between a bot that occasionally invents a return policy and one that quotes your real one.

Under the hood, every document you upload — PDFs, Word docs, spreadsheets, or plain URLs — is split into smaller chunks, converted into vector embeddings, and stored so they can be searched by semantic similarity rather than exact keyword matches.

// Simplified: how a JestBot reply is grounded in your data
const chunks = await ragService.retrieveRelevantChunks(botId, userMessage);
const reply = await llmService.generate({
  systemPrompt: bot.instructions,
  context: chunks,
  userMessage,
});

Because the retrieval step happens on every message, the bot's answers stay current with whatever is in the knowledge base at that moment — not whatever was true when the model was last fine-tuned. If you upload a corrected pricing sheet at 10am, every channel is quoting the new price by 10:01.

One bot, every channel

Create the bot once in the dashboard, and turn on the channels you need without retraining anything:

• Website Widget — a lightweight embed script for your site
• WhatsApp — connected through the WhatsApp Cloud API via embedded signup
• Mobile SDK — for a native Android or iOS app, including a packaged APK
• AI Calling Agent — for inbound and outbound phone conversations

Every channel shares the same document set, the same tool definitions, and — where the customer is identifiable — the same conversation history. A returning customer who chatted on the website last week and calls in today doesn't have to start from zero.

Tool calling as a first-class citizen

Answering questions is table stakes. What actually moves the needle for a business is a bot that can take action: check real inventory, apply a discount code, create a support ticket, or book a slot on a calendar. JestBot treats tool calling as a core capability rather than an add-on, which is what turns a chatbot into an actual automation layer for the business, not just a smarter FAQ page.

A day in the life of a connected bot

It helps to walk through what this looks like in practice. A visitor lands on your pricing page and asks the website widget a question about your Pro plan. The bot answers from your live pricing document, and — because it's a genuine lead — logs it as one. That evening, the same person messages your WhatsApp number asking whether the plan supports a feature they need; the bot recognizes the returning lead, answers from the same knowledge base, and offers to book a demo. Two days later they call your sales line; the AI calling agent already knows they were quoted the Pro plan and asked about that specific feature, so the conversation picks up naturally instead of starting over. Nobody had to manually pass notes between channels — it was the same bot the whole time.

Security and data handling

Every bot's knowledge base, conversation logs, and tool credentials are scoped to your account and, where relevant, your team. Integration credentials for things like WhatsApp, payment providers, or your CRM are stored encrypted, and access to sensitive tools can be restricted per agent or per team so a support rep and an admin don't have the same blast radius if something goes wrong.

Who this is actually for

We built JestBot for teams who are tired of maintaining five disconnected tools to do one job: answer customers and, increasingly, close sales. That includes SaaS companies who want to reduce churn with better in-app support, coaches and creators who want to scale their expertise without hiring a team, agencies who need to support dozens of clients without linearly scaling headcount, and e-commerce brands who want a bot that can actually sell, not just answer FAQs.

Comparing the cost of five tools versus one bot

It's worth putting real numbers next to the fragmented-stack problem described earlier. A separate website widget vendor, a WhatsApp reseller, a calling/IVR platform, and the internal time spent keeping all three in sync each carry their own subscription cost, their own support contract, and their own onboarding curve for whoever on your team has to manage them. Beyond the subscription fees, there's a harder-to-see cost: every hour spent updating three different dashboards with the same pricing change is an hour not spent on the product or the customer relationship itself. Consolidating onto one bot doesn't just reduce the number of invoices — it reduces the number of places a mistake can happen.

How teams actually roll this out

The teams who get the most out of JestBot don't try to turn on every channel on day one. A typical rollout looks like this: start with the website widget, since it's the fastest channel to launch and the easiest to correct if the bot gives a wrong answer. Spend a week or two watching real conversations, refining the knowledge base and instructions based on what customers actually ask (which is almost always different from what you assumed they'd ask). Once the widget is answering confidently and correctly, connect WhatsApp through the embedded signup flow — the bot is already trained, so this step is mostly just flipping a switch. From there, tool calling gets layered in once the conversational answers are solid, and the calling agent and mobile SDK come last, once you're confident the bot's underlying knowledge and behavior are dialed in. Each step builds on the last instead of trying to get five things right simultaneously.

Frequently asked questions

Do I need to set up each channel separately from scratch? No — you create and train one bot, and each channel (widget, WhatsApp, SDK, calling agent) is a toggle on top of that same bot, not a separate configuration.

What happens if I only need one channel today? Start with just the website widget or just WhatsApp — you're not required to turn on every channel. The other channels are there when you're ready to expand, using the same bot you already trained.

How current does the bot's knowledge stay? As current as your uploaded documents. Retrieval happens on every message, so there's no re-training delay between updating a document and the bot reflecting it.

Is there a right order to turn channels on in? Most teams start with the website widget, move to WhatsApp once answers are reliable, and add tool calling, the calling agent, and the mobile SDK as the bot's core knowledge and behavior get more refined.

Where this is going

The rest of this blog covers each of these pieces in depth — the WhatsApp and widget setup, the mobile SDK and APK path, tool calling and selling through the bot, and the AI calling agent for phone conversations. If you're evaluating JestBot for the first time, the short version is this: one bot, one knowledge base, every channel your customers actually use, and the tools to let it do real work instead of just talking about it.