AI chatbot development: chatbots and in-app assistants that answer from your own data

We build AI chatbots that answer from a business's own knowledge, keep to the topics it allows and pass the conversation to a person when they should. ConferBot, a no-code chatbot platform we built for its owners, is live at conferbot.com: flows are built on a visual canvas, published to a website widget and messaging channels, and every chat lands in one shared inbox. Inside two of our own products, neither open to paying customers yet, LeafRanger's Herb answers plant questions from catalogue fact cards and AskForth's Forth builds forms from a conversation.

  • Businesses answering customers on their website, WhatsApp and other messaging apps
  • Products adding an assistant that knows each user's own data
  • Teams that want an assistant's cost per turn known before it launches
21
AI model providers a ConferBot bot can run on, counted on its site
131
specialist copilots across 20 domains in Copilotly
~$0.01
per Herb chat turn on the larger model (metered in development)

What makes AI chatbots hard, and how we handled it

Answers from your data, not the model's memory

Herb answers from fact cards generated from LeafRanger's catalogue of 381,479 species, 5,071 with verified care profiles, and when a fact is missing it says so. Pet toxicity is only imported from published records, never written by a model. ConferBot's AI agent answers from the business's own help articles and documents, inside guardrails on the topics it may and may not discuss.

Handing over to a person at the right moment

ConferBot stores conversations from every channel in MongoDB and pushes them over WebSockets into one inbox, where a person can take over any chat with its full history. The AI agent escalates on negative sentiment, low confidence, a turn limit or a time limit, and passes on the transcript and any lead details it has collected.

Paying for the larger model only when a turn needs it

Herb sends greetings and simple questions to a small model and switches to a larger one when a turn carries a photo, uses words such as yellow, rot or pest, asks two or more questions, runs past 220 characters or comes 12 or more turns into a thread. A turn on the larger model cost about $0.01 in July 2026 development metering, which is why Herb is a paid-plan feature.

Conversations that change things, one undoable step at a time

Forth, AskForth's assistant, first classifies each message as build, edit or chat, and every change it makes lands on the form as one undoable step, within a 60-second deadline at temperature 0. AlwaysLoud's Reelberg edits a video storyboard only through typed operations, the same ones its buttons and its nine-tool MCP server use. Both products are built and not yet launched.

Case studies

ConferBot home page hero with the headline about building chatbots, channel icons and Start Building For Free buttonFeatured
Ai Mlfor ConferBot

ConferBot — No-Code AI Chatbot Builder SaaS

ConferBot is a no-code AI chatbot platform we built. Businesses design conversation flows on a visual canvas, add an AI agent that answers from their own knowledge, and publish one bot to their website, WhatsApp, Messenger, Instagram and other channels.

ReactNode.jsMongoDBNLP+2
21AI model providers a ConferBot bot can run on
Case study
LeafRanger home page hero with headline 'Your plants have an expert now' and a Herb chat cardFeaturedLaunching
Mobile Appfor LeafRanger

LeafRanger — AI Plant Care App and Plant Marketplace

LeafRanger is CountryDevs' own AI plant-care app. We built the Android and iOS app, two Node.js backends, the admin and vendor consoles and leafranger.com: photo plant ID and diagnosis, light-mapped rooms, weather-aware watering and a multi-vendor plant shop.

React NativeTypeScriptRedux ToolkitKotlin+11
~$0.004Measured cost of one AI vision scan
Case study
Copilotly home page hero with the headline 'Ask a marketer. Then ask 19 other experts.' and a Marketing and Growth copilot sample cardFeatured
Chrome Extensionfor Copilotly

Copilotly — Specialist AI Copilots in a Chrome Extension

Copilotly is an AI assistant we built as a Chrome extension and web app: 131 specialist copilots across 20 professional domains, from legal and health to finance and careers, on one subscription and working on whatever page the user already has open.

ReactTypeScriptChrome APIsNode.js+2
131Specialist AI copilots in the catalogue
Case study
Form editor with the Forth AI drawer open on the right showing suggested prompts and a message box.FeaturedBuilt
Saasfor AskForth

AskForth — AI Form Builder SaaS

AskForth is CountryDevs' own form SaaS. People build a form by describing it to an AI, by drag-and-drop, by mixing both or from a template, and all four produce the same JSON. Forms run as hosted links or CSP-safe embeds, with 24 industry packs on top. Built, not yet launched.

JavaScript (ES modules)TypeScriptViteSortableJS+13
68Field types in the form engine
Case study
Video storyboard studio with stage, inspector and shot timelineFeaturedBuilt
Saasfor AlwaysLoud

AlwaysLoud — AI Content Marketing Platform for SaaS

AlwaysLoud is CountryDevs' own content-marketing autopilot for SaaS companies. It reads a company's site and app, then writes blog posts, case studies, social posts, newsletters and narrated feature videos, and holds any sentence about the business the company never published. Built, not launched.

Next.js 15ReactRedux ToolkitTiptap+15
701Server tests passing at the 17 October 2026 build
Case study

What we build

  • No-code flow builders on a visual canvas with live preview
  • One bot published to a website widget and messaging channels
  • A shared inbox with human takeover over WebSockets
  • Answers grounded in your documents, catalogue or knowledge base
  • Per-turn model routing, fallbacks and daily message caps
  • Assistants that act through typed, undoable operations and MCP tools

Stack we use

  • React
  • TypeScript
  • Node.js
  • Express
  • MongoDB
  • WebSocket
  • Hosted AI models
  • Chrome APIs
  • Model Context Protocol

Related services

Not ready for a full build? Start with a fixed price

  • Discovery Sprint₹1,50,000 / $2,900Scope, prototype and a fixed-price quote in two weeks
  • Maps & AI Cost Audit₹45,000 / $950Cut your Google Maps and AI API bills, with the fixes ranked by saving

All fixed-price offers

AI Chatbot Development: questions we get asked

Should a chatbot run on one AI model or several?

Often more than one. Herb answers simple turns on a small model and moves photos, symptoms and multi-part questions to a larger one, which cost about $0.01 a turn in development metering. ConferBot makes the model provider a setting on each bot, so moving to another model does not mean rebuilding the conversation flow.

Should we build one general chatbot or several narrow ones?

It depends on the questions. Copilotly, which we built, went narrow: 131 specialist copilots across 20 domains, each set up for one kind of problem such as a lease clause or a lab report, so users describe their situation instead of writing a prompt. The copilots in legal, health and finance carry a professional disclaimer.

What should a chatbot do when its AI provider fails?

Keep answering, more simply. If a call fails, Herb falls back to the cheaper model and then to a fixed answer, so the chat never shows an error. In AskForth's Forth, a model reply that does not parse gets one stricter retry, and the repair loop is capped at two rounds so it never fails the build.

How do you cap chatbot messages on each plan?

In code, from one tier table. Herb is not on LeafRanger's free plan; Gardener allows 50 messages a day and Botanist 200, though paid plans are not switched on yet. At the cap the API returns a 429 with an upgrade message built from the same table, so the copy cannot quote a stale number, and quotas reset at midnight UTC.

Have a project like these?

Tell us what you are building. We reply with a scope, a timeline and a fixed price, usually within 48 hours.

Start a project