RAG chatbot pricing online runs from a few thousand dollars to six figures, and most of the spread comes from what is being priced. Some quotes cover a demo wired to a model. Others include the data clean-up, the evaluation set, the guardrails and the handover. This page explains what our number covers, what changes it, and what you will still pay for after launch.
What the $8,000 covers
The AI Assistant — RAG in 30 Days package is a fixed-scope build that includes:
- Ingesting, cleaning and embedding your documents
- A retrieval pipeline with evaluations and guardrails
- A chat widget or API on your branding
- Monitoring, documentation and deployment
Every answer cites the document it came from. You own 100% of the source code and IP on delivery, with a deployment runbook. For what each of those stages does technically, read custom RAG development: what a production build includes.
What moves the price of a RAG chatbot
These adjustments come straight from our AI development cost calculator, applied to the $8,000–$11,000 base:
| Factor | Adds to the build | Adds to the timeline |
|---|---|---|
| Several data sources needing cleanup | $1,000–$1,500 | About 1 week |
| Scanned PDFs, contradictory versions | $3,000–$5,000 | About 2.5 weeks |
| Used by your customers, not just staff | $1,000–$1,500 | About 1 week |
| Public, at scale | $2,000–$3,500 | About 2 weeks |
| 1–2 integrations with your systems | $1,000–$1,500 | About 1 week |
| 3+ integrations or a legacy system | $2,000–$4,000 | About 2.5 weeks |
| Customer-facing answers need guardrails | $1,000–$1,500 | About 1 week |
| Regulated or financial subject matter | $2,500–$4,500 | About 2.5 weeks |
| Nobody technical to take it over | $500–$1,000 | Half a week |
Data is the biggest single driver. Preparation is usually 30–50% of an assistant build: scanned files, three versions of a policy, nothing marked current. Integrations are next, because every legacy system adds a week nobody planned for.
Two worked examples
Internal assistant over a messy document set. Staff ask questions about policies stored as scanned PDFs with conflicting versions. It connects to one or two internal systems, and a person checks anything important before acting on it. Calculator range: $12,000–$17,500, about eight weeks.
Customer support assistant over a clean help center. One well-kept source, no integrations, your team maintains it, but answers go straight to customers so the guardrail and evaluation work is real. Calculator range: $10,000–$14,000, about six weeks.
Both are starting points for a scoping call, not quotes. The fixed price comes after we have seen a sample of the real documents.
Running costs are separate
Model usage, hosting, storage and any third-party services are not part of the build fee unless the quote says so. To estimate them, work out roughly how many questions per month the assistant will get, how long the documents and answers are, how much data it holds, and what support you want. A fixed build price does not mean free operation, and a good quote lists these lines explicitly.
Off-the-shelf chatbot builder or a custom build?
If a hosted chatbot builder answers your real questions correctly from your real documents, use it. A custom build earns its price when you need to choose how documents are split and retrieved, measure accuracy against your own questions, control who can see which document, or run it on infrastructure you own. Off-the-shelf tools give you a generic pipeline you can't tune.
Whichever route you take, test it the same way: twenty real questions from the people who will use it, answered correctly by a person first. That afternoon of work is what separates a chatbot that survives real data from one that only demos well. Why AI projects fail explains the pattern, and RAG vs fine-tuning covers why RAG is usually the right starting point.
Where this comes from
Retrieval and generation working together in production is the core of Verita AI, a voice-first clinical platform I built that combines AssemblyAI speech-to-text with Claude, Groq and Gemini for entity extraction, multi-turn chat and consultation summaries generated from the visit itself. For the full picture across all three packages, see how much it costs to build an AI app.
Frequently asked questions
How much does a RAG chatbot cost?
At Kortex Labs a production RAG chatbot over your own documents starts at $8,000, fixed price, for about four weeks of work. The calculator base range for an internal assistant over clean documents is $8,000 to $11,000. Messy data, customer-facing use, integrations and regulated subject matter add to that.
How long does it take to build a RAG chatbot?
About four weeks for a scoped assistant over clean documents. Scanned or contradictory documents add roughly two and a half weeks, and each set of integrations adds one to two and a half weeks.
Are model and hosting costs included?
No, not unless the quote says so. Model usage, hosting, storage and third-party services are separate from the build fee and depend on question volume, document size and support needs.
Who owns the chatbot after it is built?
You do. 100% of the source code and IP transfers on delivery, with documentation and a deployment runbook, so any other engineer can run it.