AI Platform
The layer your bots, assistants and agents run on: a model, your knowledge base and a place to run them, chosen by your requirements.
Your chatbots, assistants and AI agents all run on one layer: a language model, your knowledge base and somewhere to run them. Our AI infrastructure consulting starts from your requirements: what your data allows, what the law in your country says, how many requests you expect and how good the answers must be. We show which option fits and what it costs each month. If your data may leave the company, that usually means the cloud; if it may not, a closed network on servers you control. Then we set it up and keep it running.
Gen 1 of 5 — The foundation for all AI systems. What does this mean?
Buying a chatbot or an agent from us that runs on our infrastructure? The AI platform is already included in its price. If your data must stay in your own country or in your own accounts, or you bring a bot, assistant or agent of your own, we set the layer up there instead and it is paid separately.
What the AI platform is
The AI platform is everything a bot or an agent needs before it can answer anyone: the model that writes the answers, the knowledge base it reads from, and the hosting and monitoring around them. Preparing your documents and catalog for that knowledge base belongs to the bot or agent project.
We don't sell a platform licence. When you buy the AI platform on its own, we set it up in your accounts or on your servers, and you keep it; this standalone project is offered on the method we run for our own engine, and we have not delivered it for a client yet.
Your data decides where it runs
| Your requirement | What fits |
|---|---|
| Not sure yet | Start in the cloud with a pilot on your own documents and questions, and decide on the result |
| Your data may leave the company but must stay in your country | A cloud model hosted in your jurisdiction |
| Your data must not leave the company (trade secrets, an internal rule) | Private LLM deployment: open-weight models on your own servers. If your security team approves a hosting provider under contract, a dedicated server there is the alternative; the data then sits with that provider |
| Your customers' personal data is in the requests | A cloud model in your jurisdiction; or a closed network; or a foreign model behind a privacy gateway that removes the personal details it detects before a request leaves and restores them in the answer. The gateway reduces what the outside model sees; it is offered, not yet built |
| You have no IT team to run it | The cloud, with our monthly support |
A closed network is chosen because your data may not leave the company, not to save money. On the chatbots our engine runs, mostly scripted dialogues, the model's usage is a small part of the cost next to our setup and support fees (83 chats measured, July–October 2026), so a server of your own pays off only at very high volume. Agents that read many documents per task use far more, and the sizing spec runs the numbers for yours.
Prices when the AI platform runs in your own accounts or on your own servers (in a chatbot or agent from us on our infrastructure it is included; we don't resell hardware or licences):
- Cloud: setup from $3,000, support from $500 a month.
- Closed network (a private deployment on servers you control): setup from $10,000, remote support from $1,000 a month. Offered: we have not installed a closed network for a client yet.
Looking for a chatbot for your customers or an assistant for your staff rather than the layer under it? See AI chatbots and AI assistants. For AI agents built by us, the platform comes with the agent. Setting up AI coding for your software team is a separate service: vibe coding for software teams.
What You Receive
A written recommendation
Your requirements on paper, the option that fits them, one or two alternatives, and the monthly cost of each.
A model chosen on your questions
We test two or three candidate models on 20–30 of your own questions and keep that test set, so any later change of model is tried on the same questions first.
Your knowledge base, searchable
The infrastructure your bots and agents read your documents from, set up so an answer can show the document it came from.
Set up in your name
We open the cloud account in your company's name or, for your own servers, write the configuration for the provider you choose and install it. Request logging is switched off where the vendor allows it.
Monitoring and a monthly note
We watch the AI platform's cost and answer quality and send you a short note each month on what changed.
Connected to what you run
The AI platform connected to your bots, assistants, agents and the systems they work with.
How a Platform Project Goes
-
1
Free call
30 minutesWe talk through what you run or plan to run, then send a short questionnaire about your data and your volume.
-
2
Sizing spec
A written document: two or three models tested on 20–30 of your questions, the recommended setup with alternatives and monthly costs, and the pilot plan. Required for a closed network, optional for the cloud.
-
3
Pilot
about 2 weeks (cloud) · 3–4 weeks (closed network)The AI platform runs on your own documents and questions; a closed-network pilot uses non-sensitive or synthetic documents. The pilot is part of the setup price. The next payment is due only when the pilot meets the criterion set in the spec.
-
4
Production
The AI platform goes live under your bots, assistants and agents.
-
5
Support
MonthlyMonitoring, knowledge-base updates, answer quality re-checked after each change, and a model switch when a better fit appears, which we propose and you approve.
What stays with you: the accounts, the knowledge base, the test set of your questions, the documentation and the runbook.
What we don't do: sell or resell hardware or licences (you pay vendors and hosting providers directly); build a platform product for you to resell; promise that any setup makes your data handling legal.
What we have run so far
Our own chatbot engine runs on this kind of layer: one shared model and knowledge-base setup serving chatbots on seven live sites, four of them other companies' (October 2026). Request logging is switched off on every model call it makes, and an answer in our website chat can show the pages it came from. So far we have built this layer inside our own engine, on one cloud model provider. A standalone platform project in your own account, moving an existing bot, a closed-network install and the privacy gateway are offered on the same method; we have not delivered them for a client yet. The sizing spec and the pilot are there so you can check the fit before you pay for the rest.
Frequently Asked Questions
Cloud or closed network, which one do we need?
It depends on your data, not on the technology. If your data may leave the company, the cloud is usually faster and cheaper to start. If it may not, you need a closed network on servers you control. If you are not sure, start with a cloud pilot on your own documents and questions. The sizing spec writes the answer down for your case.
What do you need from us?
We need answers to a short questionnaire about your data, the rules you work under, the number of requests you expect and how good the answers must be. We also need 20–30 real questions your staff or customers ask, and a person who can approve access to your cloud account or servers.
How long does it take?
A cloud pilot takes about two weeks and a closed-network pilot about three to four, counted from the sizing spec or, on the cloud path without one, from the start of setup. We confirm the dates when we agree the work.
What does the AI platform cost?
If your chatbot or agent from us runs on our infrastructure, the AI platform is included in its price. On your own infrastructure, a cloud setup starts from $3,000 with support from $500 a month, and a closed network from $10,000 with remote support from $1,000 a month. The $1,500 sizing spec counts toward the setup. You pay the vendors directly for model usage, cloud resources and servers.
Who owns the accounts and the setup?
You own the accounts and the setup. The cloud account is in your company's name, the servers are yours or rented by you, and the knowledge base, test set, documentation and runbook stay with you if you stop working with us.
What if a vendor changes its prices or terms?
We keep the choice of model switchable. Before any switch we run your test questions on the alternative, and we switch only after you approve the result. Changing the model that indexes your knowledge base means re-indexing it; we tell you that up front.
How is the sizing spec different from your AI consulting assessment?
The sizing spec is about infrastructure: which model, where it runs and what it costs a month. The AI consulting assessment is about use cases: where AI pays off in your business and in which order.
We already have a chatbot. Can you move it?
If you control the model your chatbot calls, yes: we move the AI platform behind your chatbot to the setup that fits, priced as a platform setup (from $3,000 in the cloud). We have not moved a client's bot yet, so the sizing spec checks the move before you pay for it. A bot built inside a SaaS bot builder usually can't be moved, because the builder owns that layer.
Price
Sizing spec
$1,500
paid upfront; counts toward the setup
- Your requirements on paper
- 2–3 models tested on your questions
- Recommended setup and alternatives with monthly costs
- The pilot plan
Cloud
from $3,000
AI platform setup · support from $500 / mo
- Model and cloud chosen on your questions
- Knowledge-base infrastructure
- Pilot of about 2 weeks
Closed network
from $10,000
AI platform setup · remote support from $1,000 / mo
- Sizing spec required
- Open-weight models on your servers
- Pilot of 3–4 weeks on non-sensitive documents
- Offered: not yet installed for a client
You pay Ksentra for the sizing spec, the AI platform setup and support. You pay the vendors and hosting providers directly for model usage, cloud resources and servers. If your chatbot or agent from us runs on our infrastructure, the AI platform is included in its price. We don't resell hardware or licences.
Every AI platform project starts with a free 30-minute call.
Want your bots and agents on a layer that fits your data?
Tell us what you run or plan to run, and we will show which option fits your data and what it costs a month.