CHATS · AGENTS · AUTOMATION
AI implementation for business
We build AI chats, agents and automations that work in your systems: CRM, calendar and email. We start with a paid audit and with an answer to whether your process needs AI at all.
A fixed price after the audit, before we start building. Code, model instructions and the knowledge base go to your repository after every project stage. A “this is AI” notice and handover to a human are part of every implementation.
- REQUESTS TO THE CHAT SERVER BEFORE A CLICK
- 0
- WIDGET AFTER COMPRESSION, LOADED ON CLICK
- 5.8 kB
- SITUATIONS IN WHICH IT HANDS OVER TO A HUMAN
- 6
- AFTER HOURS IT OFFERS A SLOT FROM THE CALENDAR
- 24/7
AI FOR BUSINESS · 4 SITUATIONS
AI for business makes sense in four situations
AI makes sense in a business when customers ask about what is already on your website, an inquiry sent on Friday evening waits until Monday, someone retypes text from emails into a system, or every case needs a decision with exceptions. Each of these situations leads to a different solution.
SIGNAL · THE SAME QUESTIONS
Customers ask about what is already on your website
Symptom: 20 emails a week about price, timing and scope, while the answers are in your price list and service pages.
Solution: a website AI chat that answers from your content.
SIGNAL · AFTER-HOURS INQUIRIES
An inquiry sent Friday at 7 p.m. waits until Monday
Symptom: a customer writes in the evening and by morning is already talking to a competitor.
Solution: a chat with CRM and calendar that qualifies the inquiry, saves the contact and books a call in a slot that is really free.
SIGNAL · RETYPING
Someone retypes text from emails into a system
Symptom: orders from emails go into the ERP by hand, 2–3 hours a day.
Solution: automation with AI. The model reads the data, and a predefined workflow carries out the remaining steps.
SIGNAL · DECISIONS WITH EXCEPTIONS
Every case is a little different
Symptom: you handle complaints under 40-point terms, and each one means looking into 2–3 systems.
Solution: an AI agent that gathers the data and proposes a decision, which you approve.
FIRST STEP · AI AUDIT · 5 DAYS
AI audit — the first step
For a company that wants to use AI but does not know where to start.
FIXED PRICE · CREDITED TOWARD AN IMPLEMENTATION WITHIN 60 DAYS
An AI audit for your company
We start with a 2-hour workshop with your team, remotely or at your office. Then we review up to 3 processes and give one recommendation for each: a chat, automation, an agent or nothing.
AFTER 5 WORKING DAYS
What you get
- A recommendation for each process
- A specification of the first implementation with a fixed price and timeline
- An estimate of the monthly cost: tokens, hosting and support
If you order the implementation within 60 days, we deduct the cost of the audit from its price.
// AI IMPLEMENTATION VARIANTS
Four AI implementation variants
We most often build four AI solutions for business: an AI chat on your website, a chat with CRM, automation with AI and an AI agent. We give the exact price after the audit, before the contract is signed.
WEBSITE AI CHAT
A chat that answers from your content
Who it is for: a company whose customers ask about what is on its website.
Included
- A knowledge base built from your published pages, updated with every publication
- A “this is AI” notice
- Handover to a human
- A token budget
- 4 weeks of conversation review
An assistant for employees that answers from company documents is priced after the audit.
AI CHAT WITH CRM · HOW VI WORKS
A chat that saves leads and books meetings
Who it is for: a company where a conversation should end with a CRM entry and a date in the calendar.
Included
- Everything in the website chat
- Inquiry qualification
- Saving the contact and the sales opportunity in your CRM
- Booking slots from the calendar
- An inbox for your team
AUTOMATION WITH AI
A workflow with a model in one step
Who it is for: a process that can be written as a rule, except that one step requires reading text, such as data from an email.
Included
- 1–3 integrations
- Monitoring
- Documentation
- 30 days of support
The same as in our automation offer.
AI AGENT
An agent that prepares a case for a decision
Who it is for: a process with exceptions, in which someone reads emails and checks 2–4 systems.
Included
- A specification
- A test on 20–50 of your tasks
- Tools with permissions
- Human approval
- Limits
↳ Are you building an application with us and want AI features in it, such as summaries or ticket classification? Estimate it in the web application cost calculator.
METHOD · SITUATION → SOLUTION
Off-the-shelf chat, a ChatGPT subscription or a custom build — how we choose
For writing and summarizing, a ChatGPT, Copilot or Gemini business plan is enough; for simple questions without a CRM, an off-the-shelf chat on a subscription; we build an implementation when the conversation should end with a CRM entry, one step of the process requires reading text, or every case has exceptions. On the first call we hold your process up against this table. Sometimes the best answer is a solution we do not build.
Your situation | What we choose | Upfront cost (net) | Monthly cost |
|---|---|---|---|
The team wants to write, summarize and search faster | A ChatGPT, Copilot or Gemini business plan | PLN 0 setup | per-user subscription |
Simple FAQ questions, up to 300 conversations a month, no CRM | An off-the-shelf AI chat on a subscription | self-service setup | approx. PLN 200–500 |
The conversation should end with a CRM entry and a date in the calendar | AI chat with CRM | after the audit | tokens, hosting and support |
The rule can be written down, one step requires reading text | Automation with AI | after the audit | tokens, hosting and support |
Every case has exceptions and needs 2+ systems | AI agent | after the audit | tokens, hosting and support |
We compare the assistants' business plans in ChatGPT Business, Copilot or Gemini for business. Where to start with AI, what it costs and what the law says: AI in business.
// OUR OWN IMPLEMENTATION · LIVE SINCE OCTOBER 2026
Vi — the AI chat we built for ourselves, live on our Polish site
Vi is the “AI chat with CRM” variant on the Polish version of our own website. We will add the first results — the share of conversations handed over to a human and of booked meetings — after 4 weeks of observation (November 2026).
ANSWERS FROM SITE CONTENT
It knows only what we have published
The knowledge base is our published Polish articles, pages and glossary entries. It updates with every publication, and drafts never reach the chat. A test question about prices on 12 October 2026 returned amounts matching the calculator on the site.
QUALIFIES AND HANDS OVER
It saves the lead in the CRM and books a slot
A cheaper model (Claude Haiku) extracts the project type, scope, budget and timing from the conversation. The chat creates a contact and a sales opportunity in the CRM with a note. It hands the conversation over to a human in 6 situations, and outside working hours it offers a slot from the calendar.
DOES NOT SLOW THE SITE DOWN
It loads only after a click
The page loads only the button. Measured on 13 October 2026: 0 requests to the chat server before a click, a 5.8 kB widget after compression. The monthly token budget works as a brake: once it is exceeded, Vi sends a pre-written message without the model.
INCLUDED · 4 SPECIFICS
What you get in every implementation — included, not as an extra
OWNERSHIP
Code after every project stage
The code repository, model instructions, tool definitions and the knowledge base are handed to you after each of the 3 payment stages. You can switch the model in the configuration without rewriting the implementation.
TRANSPARENCY
A “this is AI” notice and handover to a human
Every system we build for you tells the user it is an AI before they write their first message. Art. 50 of the AI Act requires this from 2 August 2026, at the latest at the first interaction. The user can ask for a human at any time. Details: EU AI Act for business.
COST UNDER CONTROL
Budget and limits in the specification
A monthly token limit, a limit on steps per task and a fallback message without the model. Even after an error or abuse, the token bill will not exceed the amount written into the specification.
AFTER LAUNCH
30 days of support and 4 weeks of conversation review
For the first 4 weeks we read the conversations and action logs every week. We check where the AI got it wrong or found no answer, and we fix the knowledge base and instructions.
// PROCESS · 5 STAGES · PAYMENT 30/40/30
From audit to working AI — five stages
A chat launches 3–6 weeks after the audit, an agent after 5–12. Every stage ends with something to sign off, and payment is tied to stages 02, 03 and 04.
5 working days
AI audit
A workshop with your team and a review of up to 3 processes.
What you get
- A recommendation for each process
- A specification of the first implementation with a fixed price
- An estimate of the monthly cost
1–2 weeks · 30%
A prototype on your data
We run the solution on 20–50 of your real questions or tasks. At first, the tools only read data.
What you get
- Answer accuracy
- The time and cost of one conversation or task
These are the numbers you need to decide on the next stages.
1–6 weeks · 40%
Implementation
Integrations with your CRM, calendar, email or ERP, the AI notice, limits and handover to a human.
What you get
- A working implementation on a test environment
- Documentation
- Code in your repository
1–4 weeks · 30%
Launch on part of the traffic
The AI handles part of the conversations or cases. Every week we review the logs and improve the knowledge base, instructions and tools.
What you get
- A full launch
- A written list: what the AI does well and what stays with a human
Monthly
Support
Tests before every model change, conversation review and knowledge base fixes. Support for an agent also covers reviewing its action logs.
What you get
- An implementation that keeps working when the provider changes the model
Have a process that could be your first AI implementation?
In 15 minutes we will tell you whether you need a chat, automation, an agent — or nothing at all. If it is nothing, we will say so.
↳ The call takes 15 minutes, by video or phone.
RISKS · 4 SAFEGUARDS
Four risks of AI implementation and how we remove them
RISK · A MADE-UP ANSWER
The AI makes up an answer, and the company is liable
Answers come only from published content, and no answer in the knowledge base means handover to a human. The chat makes no binding price promises beyond the knowledge base. Before launch we test this on 20–50 of your real questions; more in AI in business.
RISK · THE BILL
The bill gets out of control
A monthly token budget, a limit on steps per task and a fallback message without the model. We treat what the user writes as data, not instructions — protection against hijacking the chat with someone else's instructions (prompt injection).
RISK · THE PROVIDER
You get locked in to one provider
The code and instructions are yours, and you switch the model in the configuration. Vi has a fallback provider in its architecture. Before every model change we run tests on the same 20–50 tasks.
RISK · PERSONAL DATA
Customer data ends up where it should not
Only the data the task needs is sent to the model. The full conversation stays in the inbox on a server in the EU, and only a summary goes to the CRM. The processing region and the data processing agreement are written into the specification.
WHO IT IS NOT FOR
When AI implementation does not pay off
AI implementation does not pay off when a rule is enough, when you get up to 300 simple questions a month with no CRM, or when the data sits in spreadsheets and inboxes — then we recommend a simpler solution.
A rule is enough
If the whole process can be written as a list of “if… then…” rules, a language model only adds cost. Then process automation is enough.
Up to 300 simple questions a month, no CRM
An off-the-shelf AI chat on a subscription of approx. PLN 200–500 a month pays off sooner than an implementation from PLN 8,500.
The data sits in spreadsheets and inboxes
When your systems have no API, the AI has nothing to work with. Sort out the data and systems first — we can help as part of our technology consulting.
TECHNOLOGIES · WHAT EACH IS FOR
What we build on
We choose the technology for the process. Most often we use:
- n8n — when your team should see the workflow and change simple steps itself.
- Node.js — when you need full control and automated tests.
- Claude and OpenAI models through the API — you pay for actual use and can switch provider.
- Infrastructure in the EU — the application, knowledge base and conversation logs stay in the EU.
FAQ · 6 QUESTIONS ABOUT AI IMPLEMENTATION
Questions before an AI implementation
What do I get in an AI audit?
The audit takes 5 working days. It includes a 2-hour workshop, a review of up to 3 processes, a recommendation for each (a chat, automation, an agent or nothing) and a specification of the first implementation with a fixed price.
If you order the implementation within 60 days, we deduct the cost of the audit from its price.
How much does AI implementation cost, and how much per month?
We give the implementation price after the audit, before we start building, and it does not change along the way.
Each month you pay for tokens by usage, hosting and support. An estimate of these costs for your scale is part of the audit.
How long does AI implementation take?
The audit takes 5 working days. After that, automation launches in 2–4 weeks, a chat in 3–6 and an agent in 5–12, depending on the number of systems to connect. You get the stage dates in the specification, before you sign the contract.
Who owns the code, instructions and knowledge base?
You do. The code, model instructions, tool definitions and knowledge base go to your repository after every project stage, i.e. 3 times during the implementation. The model belongs to the provider and you pay it for usage — you can switch it without rewriting the implementation.
Where does the data go, and what about GDPR?
We keep the application, knowledge base and conversation logs on our own infrastructure in the EU or on your server. Only what the task requires is sent to the model, and the processing region is written into the specification. We sign the data processing agreement together with the implementation contract.
Does a customer have to know they are talking to an AI?
Yes. From 2 August 2026, Art. 50 of the AI Act requires that the user knows this at the latest at the first interaction. Before the first message, Vi shows the notice (in Polish) “Vi is an AI assistant. You can ask to talk to a human at any time.” We describe the deadlines and obligations in EU AI Act for business.
RESOURCES · BEFORE WE TALK
Before we talk
ARTICLE
AI in business — where to start, what it costs and what the law says
AI in business without the hype: how many Polish firms use it, when an assistant is enough, what it costs and what the AI Act and Polish act require.
ARTICLE
RAG AI — how AI answers from your company's own documents
RAG AI lets a language model answer from your company documents.
ARTICLE
AI chatbot for website — when it makes sense and what it costs
AI chatbot for website: when it makes sense, what one conversation costs in zloty, what the AI Act requires it to say, and GDPR and UODO guidance for Poland.
// Next step
Let’s talk about AI in your company
Describe the process that takes the most time. On the call we tell you whether an agent makes sense there and where to start.
What happens next
- We confirm the call within one working day.
- On the call we go through the process and the data an agent would work with.
- You get a recommendation on where to start and what it costs, or an honest “not worth it”.