Short answer: a simple RAG chatbot over a small set of documents can be built for a few thousand dollars, a production-grade business assistant typically costs $12,000–$45,000, and large or fully self-hosted enterprise systems go well beyond that. Running costs are usually $200–$2,500 per month. Below is where those numbers come from and how to estimate your own.
First: what are you paying for?
A RAG (Retrieval-Augmented Generation) chatbot answers questions from your own documents and data. Under the hood it has four parts, and each one costs money:
- Data pipeline — collecting, cleaning and splitting your documents, then keeping them in sync.
- Retrieval — a vector database (such as pgvector or Qdrant) plus search tuning so the right passage is found.
- The language model — either a paid API or a model you host yourself.
- The app around it — chat interface, user logins, permissions, citations, analytics and monitoring.
Cheap chatbots skimp on parts 1 and 2. That's why they confidently give wrong answers.
RAG chatbot build costs by tier (2026)
| Tier | Typical build cost | What you get | Good for |
|---|---|---|---|
| Basic | ~$3,000–$8,000 | One data source, simple semantic search, API-based model, basic web chat | Testing the idea, small FAQ or help-centre bot |
| Production | ~$12,000–$45,000 | Several sources, hybrid search + re-ranking, citations, access control, evaluation set, monitoring | Customer support, internal knowledge assistants, compliance teams |
| Enterprise / private | $45,000–$120,000+ | Large document sets, live SQL/CRM data, multi-tenant, self-hosted model in your VPC, SSO, audit logs | Regulated industries, SaaS products, sensitive data |
These are typical market ranges published for 2026 (see ProjectCostEstimator's 2026 chatbot cost guide). Rates vary a lot by region — an experienced team in Pakistan or Eastern Europe can deliver the same architecture as a US agency for a fraction of the price.
Monthly running costs
| Cost item | Typical monthly range | What drives it |
|---|---|---|
| LLM usage (API) | $20–$1,500+ | Number of questions, answer length, model choice |
| Vector database & hosting | $50–$500 | Document volume, traffic, managed vs. self-run |
| Self-hosted GPU (if used) | Several hundred to a few thousand dollars | GPU type, hours running, autoscaling |
| Maintenance & improvements | Optional retainer | New sources, accuracy tuning, updates |
One cost people forget: keeping your content current. A RAG bot is only as good as the documents behind it, so budget a little time each month for someone to update the knowledge base.
The 5 things that really decide your price
1. Number and type of data sources
Fifty clean PDFs are easy. Ten years of scanned contracts, a CRM, a SQL database and a SharePoint site that all need syncing are not. Every extra source adds connectors, cleaning and testing.
2. How accurate it must be
A marketing FAQ bot can be “mostly right”. A legal, medical or finance assistant can't. Higher accuracy needs hybrid search, re-ranking, an evaluation set of real questions and more testing — this is the biggest difference between a $5k and a $30k build.
3. Security and hosting
Using a public API is cheapest. Running everything inside your own AWS VPC, or self-hosting an open-source model, costs more up front but keeps your data private. See our guide to private LLM deployment.
4. Users and permissions
If different staff or customers may only see certain documents, permissions must be enforced at the search level, not just in the interface.
5. Integrations
Website widget, Slack, Teams, WhatsApp, your CRM, single sign-on — each channel adds work.
How to avoid overpaying
- Start narrow. Launch with one team, one or two data sources and the 50 questions people ask most. Expand once it proves itself.
- Ask how accuracy is measured. A good developer will test against your real questions before launch. If they can't explain how, walk away.
- Pick the smallest model that works. Many business tasks don't need the most expensive model.
- Own your data and code. Avoid platforms that lock your knowledge base into their system.
- Get a fixed-scope quote after a proper audit of your data, not a guess from a sales call.
Get an exact number for your project
At IRT Solutions we build RAG chatbots for teams that need correct, cited answers and private data. Every project starts with a free audit of your data sources and security needs, followed by a fixed quote and timeline.
Frequently asked questions
What is the cheapest way to build a RAG chatbot?
Use a hosted LLM API, one clean data source and a simple vector database. That can cost a few thousand dollars to build and tens of dollars a month to run — fine for a pilot, but expect lower accuracy.
Is a RAG chatbot cheaper than fine-tuning a model?
Usually yes. RAG lets you update knowledge by changing documents, with no retraining, and it shows sources. Fine-tuning is better for changing a model's style or format, not for adding facts.
How long does it take to build a RAG chatbot?
A basic bot can take 1–2 weeks; a production business assistant usually takes 3–8 weeks, depending on data sources and testing.