AI chatbot development that answers from your own content

If customers keep asking your team the same questions, our AI chatbot development service puts a bot on your website, WhatsApp or Slack that answers from your own documents, for $2,400 fixed price, in 1 to 3 weeks. A multi-tool build with more channels and integrations is $6,500. You see a working preview within 5 business days of the deposit, or you get the deposit back. A senior engineer verifies every chatbot before it talks to a customer.

This page covers chatbots that answer. If you want a system that also acts, such as issuing refunds or updating records, read about AI agent development. Many businesses start with an answering bot and add actions later, using the same codebase.

Custom LLM chatbots from agencies usually cost $30,000 to $80,000 to build, plus monthly running costs. At the other end, a typical Fiverr chatbot gig averages a few hundred dollars and rarely includes testing or ownership. We sit between them on purpose.

What is included in the chatbot package

• A chatbot grounded on your documents and data (help center, PDFs, policies, product pages)
• One channel: website widget, WhatsApp, Slack or email
• Guardrails that keep it on topic and inside your rules
• A test suite built from your real customer questions
• A human handoff path to your team
• A usage and cost dashboard
• Deployment on your own accounts

What is not included

• Answers about information that is not in your sources. The bot says it does not know rather than guessing.
• Unlimited channels. Each extra channel is quoted in writing.
• A custom voice clone or phone-call bot.
• Legal, medical or financial advice. We can build a bot that stays on your published content and escalates sensitive topics, but we do not claim HIPAA, SOC 2 or similar compliance.
• Model fees and messaging fees (for example WhatsApp conversation charges), which you pay to the provider.

How we compare

	Tantito	Typical agency	Freelancer marketplace
Price	$2,400 to $6,500 fixed	$30,000 to $80,000 plus monthly run costs	$200 to $2,500 per gig, quality varies
Timeline	1 to 3 weeks	2 to 6 months	1 to 4 weeks
Code ownership	Yours from day one, on your accounts	Often tied to their platform	Varies, often a hosted widget
Guarantee	Preview in 5 business days or deposit back, 30-day warranty	Rarely any	Platform dispute process

How a chatbot project runs

1. Day 0: You send your sources and the 30 to 50 questions customers ask most. We agree on scope, tone and what counts as a good answer.
2. Days 1 to 5: We index your content and ship a working preview. You test it yourself and mark answers as good or bad.
3. Week 2: We tune retrieval, add guardrails and the handoff, and connect the channel.
4. Week 3 (larger builds): Extra tools, a second channel or deeper integrations, then rollout and your team handoff.

Example chatbots this package fits

• For example, a software company’s support bot that answers setup questions from its documentation and tags unresolved chats for the support inbox.
• For example, a WhatsApp bot for a dental or physiotherapy practice that answers opening hours, pricing and preparation questions, and passes booking requests to the front desk.
• For example, an internal HR policy bot in Slack that answers “how many vacation days do I have?” from the handbook.
• For example, a store assistant on a Shopify site that answers shipping and returns questions and recommends products from the catalog.

These are illustrations of scope, not past client work.

Technology we typically use

Defaults are TypeScript, the OpenAI or Anthropic APIs, retrieval over a Postgres vector index, a lightweight chat widget for the web, and the WhatsApp Business Platform or Slack API for messaging. Hosting is on your own accounts. We choose the model by testing it against your real questions, and we adapt to your stack where one already exists.

How “AI-built, human-verified” applies to a chatbot

The agents build the pipeline and the widget fast. The human step is the part that protects your brand. A senior engineer checks that retrieval returns the right passages, runs the test suite of real and adversarial questions, confirms prompts and keys stay out of the browser, and reviews how the bot behaves when it should refuse or escalate. You receive the test results, so you can see which questions pass and which are routed to a human.

The main objection: will it hallucinate in front of my customers?

Any language model can produce a confident wrong answer, so the design does not rely on the model behaving. The bot is instructed to answer only from retrieved passages and to show or log the source it used. When retrieval finds nothing relevant, it says so and offers a human. We test with questions that are deliberately outside your content, and low-confidence answers are logged for you to review. You keep control of what it knows, because the content is yours, and you can correct an answer by editing the source document.

When this is not the right fit

A chatbot will disappoint you if your documentation is thin, outdated or contradictory, because the bot can only be as good as its sources. Plan a short content cleanup before we start, and we will flag the gaps we see in the first preview. A chatbot is also not the right answer when customers mostly need account-specific actions, such as changing an order. That is an agent job. If your support volume is only a few messages a week, a well-written FAQ page may serve you better and cost nothing.

Next steps

See all packages on the pricing page, read the guarantee, or learn how we work. When you are ready, request a chatbot quote. For the cost drivers behind these numbers, read the AI agent development cost guide or try the software cost calculator. If your problem is a repetitive internal process rather than customer questions, look at business automation.