AI voice agents for customer service.
I build AI voice agents that answer your customers' calls in their own language, work from your company's knowledge and log every call in your CRM. Anything the agent should not settle on its own goes to your team with a summary. I build the integrations with your CRM, order system or shop in n8n, so the agent works inside your systems rather than next to them. A pilot starts from around €2,000 net; the final price depends on the feature set.
Customers call in languages nobody on the team speaks, outside office hours, or with a question whose answer sits in a technical data sheet. Hiring someone for every language stops paying off quickly, and the phone number is hard to take off the website because customers trust it. A voice agent takes the first contact: it answers, responds from approved knowledge and hands your team whatever it should not handle itself.
I don't sell a platform with a per-minute licence. The agent runs on your own accounts with the AI and telephony providers, and the logic and integrations run in n8n on your server. You pay the providers for usage directly, with no markup from me. The call itself is managed by my voice gateway: it handles sessions, limits and tool access, and the keys never reach the browser or the phone. I develop it from one project to the next, which is why a pilot does not start from zero.
You can hear what such an agent sounds like right now. Alicja, the voice assistant in the bottom-right corner of this page, answers questions about my services and books meetings. She runs on the same gateway and the same n8n patterns every implementation starts from.
What I build
First-line support on the phone
The agent answers on a local number, responds from an approved knowledge base and asks follow-up questions. A complaint or an unusual case gets written up and passed to your team, and the caller hears that a person will take it from there.
Conversations in the caller's language
One agent speaks Polish, English, German, French and further languages, switching to the caller's language. I test each language separately against the same scenarios before customers hear it.
Integration with your CRM and systems
I build the integrations in n8n, so the agent connects to any system with an API: HubSpot, Pipedrive, Salesforce, BaseLinker, Shopify, a helpdesk or your own system. After the call, your CRM holds a summary, the transcript and the next step.
Order and delivery status
The agent looks up an order only after verifying the caller during the conversation, for example with the order number and phone number. Access is read-only: the agent does not change orders or trigger refunds.
Lead qualification and meeting booking
The agent asks qualifying questions, assesses the enquiry and offers open slots from your calendar. n8n, not the model, checks the booking conditions, so a caller cannot talk the agent into a meeting that should not happen.
Knowledge and rules panel for your team
Your team adds and replaces documents and writes rules such as “don't say it like this, say it like that” without a developer. Every call shows which document the agent took its answer from.
Pilot: an agent on one number, with your knowledge and CRM
A fixed price for the outcome, not billed by the hour. The amount depends on the feature set: the number of languages and phone numbers, the integrations and the scope of knowledge. After the pilot you know how the agent sounds on the phone and how it answers your customers' questions before you commit to a full rollout.
- One pilot number: a new local number or a forward from your existing one
- A knowledge base built from your documents and approved by your team
- A summary and transcript of every call in your CRM
- n8n on your server, provider accounts in your company's name
Who it is for
- Companies selling in several markets whose customers call in languages nobody on the team speaks.
- Customer-service teams answering the same questions about products, shipping and returns every day.
- Companies that want a local number in each country without hiring an agent in each of them.
- Organisations where a call should end up in the CRM rather than on one person's phone.
How I run it
- 01Scope and knowledge
We agree which cases the agent handles itself and which it passes to your team. I organise your documents into one knowledge base, and you approve the content before the agent uses it.
- 02Pilot on one number
One number, one or two languages, approved knowledge and calls logged in your CRM. Before we start, we agree the test scenarios and the acceptance threshold, for example a correct answer in 17 of 20 scenarios.
- 03Testing on a real phone
You call the agent and I work your feedback into its instructions and knowledge, usually over three rounds. You judge the voice over the phone, because a phone line sounds different from a browser.
- 04Production rollout
Production numbers, more languages, the full knowledge base with passage retrieval (RAG), the team panel, monitoring, backups and automatic transcript deletion. The project ends at acceptance; ongoing technical care is optional.
Projects
A consultant's assistant at an advisory firm answers from the firm's own documents, and Alicja on this site talks to visitors, qualifies enquiries and books meetings in the calendar.
A self-updating base of more than half a million documents, hybrid search, a research agent and analysis drafts. From ~12 hours of work to ~2 minutes.
Read the case studyA voice agent on aiadvisors.pl: a voice model, a reasoning model in the background, my own gateway and more than a dozen n8n workflows. Answers from the site's knowledge, qualifies enquiries and books meetings.
Read the case studyGuides
What a minute of conversation costs, when a custom agent beats an off-the-shelf AI receptionist, and how the agent calls n8n mid-conversation.
Voicebot, voice agent or AI receptionist: which to choose and what does it cost?
An off-the-shelf AI receptionist starts at $29 a month and can be live in a few hours. A custom voice agent on GPT-Live with Twilio telephony runs at about $0.065 per minute plus roughly $14 in fixed monthly fees, but someone has to build it. Below about 50 calls a month the ready-made service wins; above 500 a custom agent pays for itself within months. In both cases bookings, CRM lookups and quotes run in n8n, which the agent calls mid-conversation.
Read the guideWhat is the n8n AI Agent and how does it work?
The n8n AI Agent is a node in which a language model receives instructions (a system message) and a set of tools, then decides on its own which tools to call to complete a task. Without tools it is just a chat. With tools, most often sub-workflows, the agent can read a document, check data in a spreadsheet and make a judgement call no plain rule can handle. Since August 2026 n8n also has a separate Agents category: agents with their own memory, channels and skills that live outside a single workflow.
Read the guideFrequently asked questions
- How much does it cost to implement a voice agent?
- A pilot starts from €2,000 net at a fixed price. The amount depends on the feature set: the number of languages and phone numbers, integrations with your CRM or order system, and the scope of knowledge. I quote the production rollout after the pilot, also as a fixed price. I don't bill by the hour.
- How much does a voice agent cost to run each month?
- The cost grows with call minutes. At 2,000 minutes a month, the technology (the voice model, the background model, the phone number and the server running n8n) comes to around €170–310 net in my calculations. You pay it directly to the providers, with no markup from me and no per-user licence. Technical care is optional.
- How long does implementation take?
- I plan a pilot on one number at about two weeks from receiving access and materials, followed by rounds of testing and fixes. Moving to production numbers takes roughly another two weeks. Account setup and number registration depend mostly on the operators.
- Which languages can the voice agent speak?
- Polish, English, German, French and other languages the voice model supports. The agent switches to the caller's language. I test every language your customers will hear separately, because pronunciation and accent vary between languages.
- Can the agent have a local number in Germany or France?
- Yes, but both countries require a company address in that country. The rule comes from the regulators (BNetzA and ARCEP), so it applies to every operator. Without a German address, the option is forwarding to the agent's number, either from your existing number or from a new German SIM card. In France, a company without a French address can only get a toll-free 0800 number, because French mobile numbers are reserved for person-to-person communication. Mobile numbers in the UK, Poland, Czechia, Slovakia and Lithuania can be registered with an address from another country, but availability differs between operators. That is why I work with several SIP operators, and numbers from different countries can all lead to the same agent. As of September 2026.
- Does the agent sound the same on the phone as on the website?
- Not quite. A phone line carries a narrower audio band than a browser, so the same voice sounds different on a call. That is why you call the agent on a real number before deciding on a rollout. The short pauses while the agent looks something up can be shortened, but not removed entirely.
- Does the caller know they are talking to AI?
- Yes. At the start of the call the agent says it is an AI assistant and, if the call is recorded, says so as well. Article 50 of the EU AI Act requires this, and I configure every agent that way. The wording of your data-processing notice is worth checking with a lawyer.
- What happens when the agent doesn't know the answer?
- It says it doesn't know, takes a description of the issue and a contact, and passes the case to your team. It answers only from approved knowledge, and every call shows which document the information came from. Your team uses that to fill gaps in the knowledge or add a rule.
- Where is the data, and who owns it?
- We open the AI and telephony accounts in your company's name, and n8n runs on your server. The accounts, data, knowledge base and workflows are yours. We choose the model provider to fit your requirements, including processing data in Europe. You decide how long transcripts are kept, and summaries can stay in your CRM. We sign a data processing agreement.
- Off-the-shelf AI receptionist or a custom voice agent?
- For simple questions and a few dozen calls a month, a ready-made service is usually cheaper and quicker to launch. A custom agent pays off when calls run into the hundreds, when it needs your own systems, or when you need several languages and numbers. The step-by-step numbers are in the voicebot cost guide.