AI agent development for business
An AI agent is an artificial intelligence program that doesn't just answer questions but also carries out actions in your systems by itself: it reads an enquiry, creates a CRM record, checks stock and drafts a quote. Saitai.lt builds AI agents with clear permissions and employee approval for important actions. Prices start at €1,490, with a timeline of 3–6 weeks.
In short
- Price
- from €1,490 (one process with actions across several systems); no VAT is added
- Timeline
- 3–6 weeks; we show you the first working version within 2–3 weeks
- Monthly costs
- support from €29/month (on a virtual private server (VPS) we provide); AI model usage charged by volume of tasks, with a monthly limit
- AI models
- OpenAI (GPT), Anthropic Claude, Google Gemini or open-source models
- Integrations
- email, CRM, WooCommerce, Google Sheets, accounting software – via API or import
- Control
- only the permissions needed, employee approval, a log of every action
- Warranty
- 12 months
- Payment
- 50% deposit, balance when the agent goes live
- Where we work
- remotely, all over Lithuania; meetings in Vilnius by appointment, arranged by phone
Who it’s for
- A sales manager reads every enquiry, retypes the details into the CRM, checks prices or stock, and only then replies to the customer.
- A single task involves several programs: email, CRM, your online store admin, Excel or accounting software.
- In peak season, customers wait days for a reply, even though most enquiries are routine.
- The process has clear rules – a price list, discount limits, delivery terms – but too many exceptions for a simple automated workflow.
- You want AI to work with your data in a controlled way: only with the permissions it's given, with an action log and employee approval.
When it’s not the right fit
- You only need to answer customers' questions, with no actions in other systems. A chatbot is enough – it's cheaper and quicker to set up.
- The process is always the same, with no exceptions, for example transferring a paid order into your accounting software. That kind of work doesn't need AI – an ordinary integration is enough.
- The systems the agent would work with have no API and no way to import or export data.
- The AI would have to make decisions about people by itself, such as shortlisting job candidates or assessing creditworthiness. The EU AI Act classes such systems as high-risk, and we don't build them: AI can prepare a draft or sort data, but a person makes the decision.
Included
What you get
Process specification and agent rules
We set out what the agent does, under what conditions, what it never does and when it turns to an employee. This specification becomes the agent's instructions and the basis for testing.
Tools for your systems
We build specific actions for the agent: looking up a customer in the CRM, creating a record, checking stock, drafting an email. Each action gets access only to the data it needs.
Knowledge base
Price lists, delivery terms, internal rules and sample replies – the agent works from these, not from general knowledge on the internet.
Employee approval
The agent prepares a quote for a customer, a price change or an entry in your accounting software as a draft, and only sends or saves it once an employee approves it. You decide which actions can happen without approval.
Action log and spending limit
You can see what the agent did, why, and which data it relied on. AI model usage costs stay within a monthly limit agreed with you.
Testing on your own cases
Before launch, we test the agent on real emails and enquiries, including unusual cases. We repeat the tests after every change.
Launch and a 12-month warranty
We show your staff how to approve the actions the agent prepares and where to find the log. For 12 months, we fix programming bugs free of charge.
Price
How much does it cost to build an AI agent?
Price from
€1,490
Building an AI agent costs from €1,490: that's the price of an agent that carries out one process across several of your systems, such as email, your CRM and your online store, with employee approval. When the agent has to handle several processes or work with systems that have no API, we price it based on the scope.
Prices include all taxes – no VAT is added (we are not VAT-registered). The final price depends on the project: we put it and the timeline in the contract before work starts.
What the price depends on
- How many processes and scenarios the agent handles, and how many exceptions they involve.
- How many systems the agent works with, and whether they have an API.
- How many actions the agent carries out by itself, and how many it only prepares for an employee to approve.
- How many documents and examples are needed for the knowledge base and testing.
- Whether the agent talks to customers directly or works only for your staff.
- Where the data must be processed: in the model provider's cloud, in an EU data centre or on your own server.
Additional costs
- We choose where to host the agent based on its size: a VPS we provide (sized to your needs) or your own server; the price depends on this too. Support on our VPS from €29/month covers hosting, uptime monitoring and knowledge base updates.
- AI model usage (API) – charged by the actual volume of tasks, within a monthly limit agreed with you. The agent calls the model several times for each task, so we set the limit based on data from the trial period.
- New processes and systems after launch – under a separate quote, agreed in writing.
Timeline
How long does it take to build an AI agent?
An AI agent with integrations takes 3–6 weeks from receiving the deposit and examples. We show you the first working version, running on your own emails or enquiries in a test environment, within 2–3 weeks.
| Agent scope | Example | Timeline |
|---|---|---|
| 1 process, 2 systems | enquiry email → CRM record → draft reply | 3–4 weeks |
| 1 process, 3–4 systems | enquiry → CRM → stock check in the online store → quote as a PDF | 4–6 weeks |
| Several processes | enquiries, order statuses and a weekly report for the director | in stages, 3–6 weeks per process |
Process
How it works
- Free assessment1–2 working days
In a 30–60-minute call, we discuss the process: who does it now, in which programs and how long it takes. We tell you whether it needs an agent or whether a simpler solution will do.
- Process specification and fixed price1–3 working days
We set out the agent's actions, permissions and approval points, and what isn't included in the first version. The price and timeline go into the contract, and we start once we receive the 50% deposit.
- Examples and test set2–5 working days
From 30–50 of your emails, enquiries or documents, we put together a test set and describe the correct result for each case. We remove any unnecessary personal data.
- First version1–2 weeks
The agent works with your examples in a test environment, but doesn't yet carry out actions in your live systems. You see its decisions, and together we fine-tune the rules.
- Integrations and testing1–3 weeks
We connect the agent to your CRM, email, online store or accounting software and give it only the permissions it needs. We test it against the test set until the results match the specification.
- Launch with approval1–2 working days
For the first few weeks, an employee approves every action the agent takes. Once the errors are gone, you decide which actions can happen without approval.
- Monitoring and warranty12 months
During the first month after launch, we review the logs and fine-tune the rules. For 12 months, we fix programming bugs free of charge.
Comparison
Automated workflow, AI agent or doing it by hand?
| Criterion | Automated workflow (no AI) | AI agent (Saitai.lt) | An employee, by hand |
|---|---|---|---|
| What it does | the same sequence of steps, following pre-set rules | understands free text, chooses actions and carries them out within the permissions given | everything, but each task by hand |
| Free-form emails and exceptions | stops or gets it wrong if the data isn't in the expected format | recognises most cases and passes unclear ones to a person | handles them best |
| Upfront cost | usually lower than for an agent | from €1,490 | recruiting and onboarding a new employee |
| Monthly costs | workflow platform subscription or hosting | support from €29/month (on our VPS); model usage charged separately | salary and employment taxes |
| Control | a notification if the workflow fails | an action log and employee approval for important actions | depends on internal procedures |
| Best for | uniform data and always the same steps | a routine process involving free text and several systems | rare, complex tasks or those that involve negotiation |
Examples
Typical scenarios
Wholesale enquiries with stock checks
- Situation
- For example, a building materials wholesaler receives enquiries by email with product lists written in free form. A sales manager looks up each product in the catalogue and checks the stock and the price for the customer's price group.
- Solution
- The AI agent recognises the products listed in the email, finds them in the catalogue, checks stock, applies the customer's price group and drafts a quote in the CRM. It flags any products it doesn't recognise, and the sales manager reviews the quote and sends it.
- Result
- A draft quote with no manual catalogue searches, a clear list of unrecognised products and every enquiry in the CRM.
Fault reports at a servicing company
- Situation
- For example, a heating equipment servicing company receives fault reports by email and through its website form. The dispatcher works out the address and the equipment, finds the customer in the system and schedules a technician's visit.
- Solution
- The agent extracts the address, the equipment and the fault description from the report, finds the customer record and suggests a free visit slot based on the technicians' calendar. If details are missing, it drafts an email asking for them, which the dispatcher reviews and sends.
- Result
- A completed job record, a suggested visit time and fewer calls to clarify details.
Weekly report for the director
- Situation
- For example, the director of an e-commerce company asks for a summary every Monday: sales, unpaid invoices and products that are running low. The accountant gathers the data from several systems.
- Solution
- Every week, the agent gathers data from the online store, the accounting software export and Google Sheets, compares it with the previous week and prepares a short report. The figures come straight from the systems, and the AI only describes the changes.
- Result
- A report on Monday morning, a clear source for every figure and more of the accountant's time for other work.
Describe the process – we'll assess whether it needs an agent
Tell us which work your staff do across several programs, and how often. Within 1 working day, we'll tell you whether an AI agent is the right fit, how much it would cost and how long it would take.
FAQ
Frequently asked questions
It can, but only for the actions you allow. At first, an employee approves every action the agent takes, and later the actions usually left to run without approval are the safe ones, such as creating a CRM record or sending an internal notification. Matters involving money, prices and contracts stay with an employee.
More
Related services
AI chatbots for websites
When answering customers based on your prices and rules is enough – from €490.
Custom skills for AI agents
Your procedures and templates in the AI assistant your team already uses – from €390.
AI integration into websites and systems
A single AI task in your website, CRM or online store – from €990.
Custom CRM development built around your processes
When the agent needs somewhere to record customers, enquiries and tasks.
Enquiry
Get an AI agent assessment within 1 working day
Describe the process you want to hand over to an agent and the programs it involves. You can email us sample emails or enquiries without personal data. Tel. +370 670 99265.
Phone / WhatsApp / Viber:
+370 670 99265
Email:
info@saitai.lt
We reply within 1 working day. Meetings in Vilnius are by appointment only – please call first.

