In this article
- What does Dante AI cost?
- Step 1: What job does your AI agent have to do?
- Step 2: How many answers do you need each month?
- Step 3: Which build path does that point to?
- Step 4: What do agency and custom builds really cost?
- Step 5: Which costs appear after you sign?
- What should you actually pay in 2026?
- What your first week tells you about the bill
How much does it cost to build a chatbot?
A chatbot costs $0 to $150,000. On a no-code platform such as Dante AI it is $0 to $400 a month and live in 60 seconds. An agency charges $2,000 to $25,000 to set up, plus a monthly retainer. Custom development costs $10,000 to $150,000 upfront, then 20 to 40 percent of that every year to keep it running.
Here is the same answer as a budget line, followed by five steps that turn your own numbers into a figure you can defend.
| No-code AI platform | Agency build | Custom development | |
|---|---|---|---|
| Setup | $0 | $2,000 to $25,000 | $10,000 to $150,000 |
| Monthly | $0 to $400 | $500 to $5,000 retainer | $500 to $5,000 running costs |
| First year | $0 to $4,800 | $8,000 to $85,000 | $16,000 to $210,000 |
| Live in | 60 seconds | 2 to 8 weeks | 3 to 9 months |
| Upkeep | Included | Inside the retainer | Yours to fund |
What does Dante AI cost?
This is a page about money, so our own prices go near the top rather than at the bottom.
- Free: $0. 100 credits every month, and up to 700 free credits during onboarding. Enough to train an AI agent on your website and judge its answers before paying anything.
- Starter: $40 a month. 2,000 credits, advanced AI models, and integrations with the tools you already use.
- Advanced: $120 a month. 8,000 credits, for an agent carrying a real share of your customer conversations.
- Pro: $400 a month. 30,000 credits, for high-volume support operations.
Credits are metered per answer, not per seat, so the bill does not climb when your team grows. Top-ups are $20 per 1,000 credits on any paid plan, and plan detail lives on the pricing page.
Price it with your own content. Train an AI agent on your website, docs or files. Live in 60 seconds. No code needed.
Step 1: What job does your AI agent have to do?
Write the job in one line. Answer questions from our website. Qualify leads and hand them to sales. Book appointments. The narrower and more standard the job, the cheaper every path becomes, because standard work is already built.
Answering from your own website, documents and help pages is the standard job, and a platform does it for a subscription. Custom conversation logic, transactions inside your own systems, or an unusual compliance regime is the expensive job, and it is the only good reason left to commission a build.
Step 2: How many answers do you need each month?
Count the customer questions you handle in a normal month. Volume is the biggest driver of ongoing cost on every path, because each answer carries a model cost behind it.
On our plans an answer costs 1 to 25 credits depending on which AI model you choose, so a plan converts straight into a number of answers. The free plan runs the 1-credit models, so 100 credits is 100 answers. Model choice opens up from Starter.
| Model choice | Credits per answer | Free ($0) | Starter ($40) | Advanced ($120) | Pro ($400) |
|---|---|---|---|---|---|
| Standard | 1 | 100 | 2,000 | 8,000 | 30,000 |
| Mid-weight | 5 | Starter and up | 400 | 1,600 | 6,000 |
| Advanced reasoning | 10 | Starter and up | 200 | 800 | 3,000 |
| Heaviest | 25 | Starter and up | 80 | 320 | 1,200 |
Find your monthly question count in the row for the model you want and you have your tier. On a paid plan you can switch model without retraining, so cost per answer is a dial you control rather than a quote you accept.
Step 3: Which build path does that point to?
A standard job at normal volume points to a no-code chatbot platform at $0 to $400 a month, live the same day. A standard job with nobody in-house to run it points to an agency, at a setup fee plus a retainer. Genuinely bespoke logic or deep integration with proprietary systems points to custom development at $10,000 and up.
The pattern worth noticing: for most businesses, growing means moving up a tier, not leaving the platform. The jump to a custom build is driven by unusual requirements, not by size. If you are weighing a large deployment against hiring, compare it with the cost of implementing AI in a call center.
Step 4: What do agency and custom builds really cost?
Custom development is a software project, so you are paying for engineers, project management, infrastructure and permanent ownership. The 2026 ranges quoted most often:
- Simple rules-based flow: $10,000 to $30,000. Scripted paths, limited understanding.
- AI assistant with retrieval: $30,000 to $80,000. A model connected to your knowledge, which is what platforms now ship off the shelf.
- Enterprise conversational system: $80,000 to $150,000 or more. Multi-system integration, custom compliance, dedicated infrastructure.
Agencies usually configure a platform for you, then charge $500 to $5,000 a month to manage it. That buys accountability and costs you speed, since every change routes through them. A common ending is that the agency stands the agent up on a platform and the team then brings it in-house, turning a retainer into a subscription.
The number on the proposal is never the whole number. Running costs commonly add 20 to 40 percent of the build price each year, and across three years they often exceed the build itself.
Step 5: Which costs appear after you sign?
These are the lines that turn a cheap headline price into an expensive year.
- Training data. Outdated help docs produce a confidently wrong agent. A platform that reads your live site takes this close to zero. Custom builds routinely spend weeks preparing content.
- Model usage. Every answer has a model cost. Platforms bundle it into plan credits. Custom builds pay the API bill directly, and it grows with traffic.
- Integrations. Connecting your help desk, CRM or booking tool is configuration on a platform and a priced line item everywhere else. Ask whether human handover is standard, an add-on, or bespoke work.
- Maintenance and retraining. Your product and policies change and the agent has to follow. Included in a subscription, 20 to 40 percent of build cost a year otherwise.
- Seat creep. Per-seat tools rebill you for every person you add, so a growing team quietly doubles the bill. Metered pricing charges for answers instead, tying cost to customers rather than headcount.
- Wrong answers. The most expensive agent is the one that answers incorrectly and loses a customer. Test it against real questions before it faces the public.
What should you actually pay in 2026?
- Testing the idea: $0. Train it on your own site and read the answers. If it cannot prove itself free, do not pay for it.
- Typical small business: $40 to $120 a month, live the same day.
- High-volume support operation: $400 a month plus top-ups, still a fraction of one support hire.
- Genuinely bespoke enterprise requirement: $50,000 and up, plus months of build time, justified only when a platform demonstrably cannot meet a hard requirement.
When people ask us for a number, the question underneath is almost always whether the answers will be good enough to trust. Price is easy to compare. Answer quality on your own content is the thing that decides whether the price was worth paying, and you can settle that today for nothing.
What your first week tells you about the bill
A week of real conversations replaces every estimate on this page with your own numbers.
- Your actual monthly question count, which is the one input that decides your tier
- The share answered from your own material without a person
- Which model you genuinely need, and therefore your real credits per answer
- Leads captured in the conversation instead of lost on a contact form
- A ranked log of what customers ask, which is the cheapest research you will get
Point it at your website, let it learn your pages, paste one script tag, then read the first week of conversations. Start free, multiply your question count by your credits per answer, and the tier picks itself on the pricing page.
Further reading
Keep going with these guides from the Dante AI library: