Bengaluru-based · Remote worldwide
AI Chatbot & AI Agent Development in Bangalore
Website chatbots, WhatsApp bots, assistants that answer from your own documents, and agents that actually do things — built by the person you talk to, with the monthly running cost quoted before anything is built.
- Website chatbots — grounded in your own pages, with a human handoff
- WhatsApp bots — on the official Business API, not a bannable workaround
- RAG assistants — answers from your documents, with citations
- Autonomous agents — book, update, escalate, with approval gates
- Workflow automation — often the cheaper half of the problem
The short answer
A starter website chatbot is ₹35,000–₹75,000, a business bot with WhatsApp and CRM integration ₹1,00,000–₹3,00,000, and a custom multi-step agent ₹2,50,000 upwards — plus ₹2,000–₹80,000 a month to run, which is the figure most quotes leave out. Live in two to four weeks for a bot; six to fourteen for an agent.
The question worth answering first is chatbot or agent. A chatbot answers; an agent does something. Agents cost several times more because every action needs error handling, an approval gate and an audit trail — so if answering is enough, I will say so and build the cheaper thing.
What I build
Five things, and the fifth is often the right one
Website chatbots
A widget on your own site that answers from your pages, pricing and policies — not a generic model guessing. Handoff to WhatsApp or a form when it does not know.
WhatsApp bots
On the official WhatsApp Business API, so it survives volume and does not get your number banned. Order status, FAQs, lead capture and appointment booking.
RAG knowledge assistants
Answers grounded in your own documents with a citation on every reply, and a refusal when no source supports it. This is the one that replaces internal Q&A.
Autonomous agents
Systems that take actions rather than answer questions — check stock, raise a ticket, book a slot, update the CRM — with a human approval step wherever money or commitments are involved.
Workflow automation
The bot is often the wrong half of the problem. Document extraction, routing and reconciliation in n8n or code, with a model only where judgement is genuinely needed.
How it actually works
A grounded chatbot is not a model with your logo on it
The model is one box in the middle. Everything that makes the difference between a demo and something you can put in front of a customer is the retrieval, the tools and the refusal behaviour around it — which is why "we'll just plug in ChatGPT" quotes come in low and land late.
Use cases
Four builds, and what each one actually removes
Support bot
A Bengaluru equipment dealer's top 20 questions — warranty terms, service turnaround, spare availability — answered from the actual service manual, with the AMC page cited. Anything about pricing on a live order goes to a human.
Deflects the repeat questions; the exceptions still reach a person.
Lead-generation bot
Qualifies a website enquiry before it becomes a lead: budget band, timeline, city, what they have already tried. Writes a structured record instead of "please call me".
Sales stops opening enquiries that were never going to buy.
Booking agent
Reads the calendar, offers real slots, books one, sends the confirmation, and reschedules when someone replies "can we move this". Takes an action, so it is an agent rather than a chatbot.
Removes the back-and-forth, not just the first reply.
Internal ops agent
Staff ask it in plain language: which invoices are unpaid past 45 days, what did we quote this client last time, where is this consignment. It queries the systems that already hold the answer.
Stops new staff interrupting senior staff to find things.
Tech stack
What these get built on.
Models from OpenAI, Anthropic and Google, chosen per job on cost, latency and data terms rather than loyalty. Retrieval over a vector store — Pinecone, or pgvector on Postgres/Supabase when you would rather not add a service. Orchestration in LangChain or plain code, workflow automation in n8n or Make, and WhatsApp on the official Business API. The model sits behind an interface so it can be swapped when a cheaper one appears.
OpenAI
Claude
Gemini
LangChain
Pinecone
Supabase
n8n
WhatsApp Business
Python
FastAPI
TypeScript
Vercel
Which one do you need
Rule-based bot, AI chatbot, or autonomous agent?
Most enquiries ask for the third and need the second. A few genuinely need the first — and if you have five questions that never change, a decision tree is the honest recommendation and I will tell you that rather than sell you a model.
| Rule-based bot | AI chatbot | Autonomous AI agent | |
|---|---|---|---|
| What it can do | Follows a decision tree you drew. Answers only what you scripted, exactly as scripted | Understands the question in the user's own words and answers from your documents, with citations | Understands, then acts — queries systems, books, updates, escalates. Chains several steps |
| When it fails | Any phrasing you did not anticipate. Users hit "talk to a human" within two turns | Questions with no source in your content — it should refuse rather than invent, which is a build decision | Multi-step tasks where one tool returns something unexpected. Needs real error handling and approval gates |
| Build cost | ₹15,000–₹40,000 | ₹35,000–₹3,00,000 depending on integrations | ₹2,50,000–₹15,00,000+, driven by how many systems it touches |
| Monthly running cost | Near zero — no model calls | ₹2,000–₹25,000, usage-based | ₹8,000–₹80,000, plus the cost of whatever it calls |
| Setup time | 3–7 days | 2–4 weeks | 6–14 weeks |
| Worth it when | You have five questions, they never change, and the budget is tiny | People ask the same things in a hundred different ways and the answers live in documents | The work is not answering — it is doing something across systems that a person currently does by hand |
The payback maths
A bot costs more than a person for the first six months
Vendor charts usually start at month twelve, where the answer is flattering. This one starts at month zero, where the build cost is real and the bot is the more expensive option. Every input is listed below — substitute your own salary figure and the crossover moves.
- Support executive
- ₹25,000/month salary — the mid of published Indian averages — loaded ×1.25 for PF, ESI, workspace, equipment and supervision, so ₹31,250/month to employ
- Business chatbot
- ₹1,50,000 to build, once, plus ₹8,000/month for model usage, hosting and monitoring
- Break-even
- Month 7. Before that the bot is behind; by month 12 it is ₹1,29,000 ahead
- The honest caveat
- A bot does not replace a person one-for-one. It absorbs the repetitive fraction, so in practice this is the cost of not making the next hire rather than of removing someone. If a vendor shows you this chart without that sentence, discount the rest of their numbers too.
Salary basis: published Indian customer-support executive averages of roughly ₹20,600–₹25,400 a month. Build and running figures are the mid of the package ranges below. Sources are listed in pricing. Your numbers will differ — the arithmetic is shown so you can redo it with them.
Indicative pricing
Packages, and what moves you between them
Indicative Indian market ranges, not a rate card. What moves a build up a band is almost always integration count rather than conversation volume — a bot that answers from documents is a fraction of the work of one that writes to your CRM, checks a calendar and raises a ticket.
| Package | Indicative price | What is included | Live in |
|---|---|---|---|
| Starter bot | ₹35,000–₹75,000 | Website widget, trained on up to ~50 pages or documents. FAQ deflection, lead capture, WhatsApp handoff. One language | 2 weeks |
| Business bot | ₹1,00,000–₹3,00,000 | Everything above plus WhatsApp Business API, CRM write-back, appointment booking, analytics dashboard, English + one Indian language | 3–4 weeks |
| Custom agent | ₹2,50,000–₹15,00,000+ | Multi-step agent across your systems: tool calling, approval gates, evaluation harness, logging and monitoring, human escalation | 6–14 weeks |
| Running cost | ₹2,000–₹80,000 / month | Model usage, hosting and monitoring. Forecast from your real volumes and capped before launch — billed to your own accounts at cost | Ongoing |
| Care plan | ₹8,000–₹40,000 / month | Optional. Tuning against logged failures, content refreshes, model migrations as prices fall, monthly report | Ongoing |
WhatsApp adds Meta's own per-conversation charges on top, billed by Meta to your account. Model and hosting costs are always billed to your accounts at cost, never resold at a margin. Every figure is sourced in pricing.
Delivery
What the four weeks actually contain
This is the business-bot timeline. A starter bot compresses it to two weeks; an agent stretches the build and integration stages considerably. You see a staging link in week one regardless.
- Week 0 — scope
- The questions it must answer, the ones it must refuse, and where the content lives. Running cost modelled against your real volumes before anything is built.
- Week 1 — evaluation set
- Fifty real questions with known-correct answers, assembled first. Without this, "is it better?" is an argument rather than a measurement.
- Weeks 1–2 — build
- Retrieval over your content, prompt and refusal behaviour, handoff rules. You get a staging link on day one of this stage and watch it improve.
- Week 3 — integration
- WhatsApp, CRM, calendar or helpdesk. Approval gates wherever the agent commits money or makes a promise.
- Week 4 — pilot
- Live to a slice of real traffic with every conversation logged. The failures from this week are the build list for the next.
- Ongoing — tuning
- Fixing what the logs show, not what we guessed. Most of the accuracy gain happens here rather than at launch.
FAQ
AI chatbot and agent development — questions I get asked
How much does it cost to build an AI chatbot in Bangalore?
A starter website bot is ₹35,000–₹75,000, a business bot with WhatsApp and CRM integration ₹1,00,000–₹3,00,000, and a custom multi-step agent ₹2,50,000 upwards. The number most quotes leave out is the monthly running cost — ₹2,000–₹80,000 depending on volume — which I forecast from your real traffic and cap before launch rather than discovering together in month two.
What is the difference between a chatbot and an AI agent?
A chatbot answers. An agent acts. Ask a chatbot to move your appointment and the best it can do is tell you how; an agent reads the calendar, finds a slot, books it and sends the confirmation. Agents cost several times more because every action needs error handling, an approval gate and an audit trail — so the honest question is whether answering is genuinely not enough for your case.
How long does it take to build?
Two weeks for a starter bot, three to four for a business bot with integrations, and six to fourteen weeks for a custom agent. You see a staging link in the first week either way. The variable is almost never the model — it is how quickly your content and system access can be assembled.
Will it make things up?
Not if it is built to refuse. Every answer is grounded in retrieved content with a citation, and where no source supports the question the correct behaviour is to say so and offer a human. That is a build decision rather than a property of the model, and it is the single thing that separates a bot you can put in front of customers from a demo.
Can it work on WhatsApp?
Yes, on the official WhatsApp Business API — which matters, because the unofficial libraries get numbers banned once volume rises. There is a Meta approval step and per-conversation charges from Meta on top of the build, and I will show you that cost against your expected volume before you commit.
Which model do you use — OpenAI, Claude or Gemini?
Whichever fits the job on cost, latency and data terms, and I will tell you why rather than defaulting to one. Most builds are not locked to a provider: the model sits behind an interface so it can be swapped when a cheaper or better one appears, which happens roughly every few months and has repeatedly cut clients' running costs.
Is my company data safe?
The business and API tiers of OpenAI, Anthropic and Google contractually do not train on your inputs and offer zero-retention options; consumer tiers are a different contract entirely. Part of scoping is deciding what may leave your network at all — and where the answer is nothing, the build uses an open-weight model on your own infrastructure, which costs more and performs less well. I would rather say that up front.
Do you work with clients outside Bangalore?
Yes. I am Bengaluru-based, so local clients can have the scoping session in their office, which genuinely helps — most of the useful detail comes from watching people handle the enquiries you are about to automate. Everything after that is remote, and chatbot and agent work is delivered worldwide.
What happens if it stops working after handover?
Everything is in your accounts and your repository from the first commit, so nothing depends on me being reachable. There is a warranty period after launch, then an optional care plan. If you would rather take it to another developer, the documentation and evaluation set go with it — that is deliberate.
Can you improve a chatbot someone else built?
Often, and it is usually cheaper than rebuilding. The common findings are the same three: no evaluation set, so nobody can tell whether changes help; retrieval that returns roughly-relevant content instead of the right passage; and no refusal behaviour, so it answers confidently when it should not. I will look at it and tell you honestly whether fixing or restarting is the better spend.
Tell me what it should handle.
One message: what the bot or agent needs to do, roughly how many conversations a month, and where the content lives. You get a written, itemised quote — including the monthly running cost — within 24 hours. If a decision tree would do the job, I will tell you that instead.
100+ clients served·Free, no-obligation quote·NDA on request·You own everything at handover