Chatbot platforms and conversational-AI agencies
Give your chatbot customers a phone line
A voice assistant on VoiceDock takes the call and asks your chatbot for the answer while the caller is on the line. The content and the customer stay yours, and we build the connection with your team.
your brand · an organisation per customer · € 0,07/min orchestration · EU-hosted · DPA included
// 1 · register your platform's MCP server oncePOST /api/v1/mcp-servers{ "name": "your-chatbot", "url": "https://mcp.your-platform.com/mcp", "headers": { "Authorization": "Bearer ••••••" }} // 2 · attach it to your customer's assistantPATCH /api/v1/assistants/{id}{ "llm_config": { "tools": [ { "type": "mcp", "mcp_server_id": "2f1c9a7e-4b3d-…", "allowed_tools": ["ask_chatbot"], "required": true }, { "type": "transfer_call", "destinations": [ … ] } ] }}Who it is for
Two kinds of partner, one voice layer
Whether you sell the chatbot platform or build on top of one, the setup underneath is the same.
chatbot and AI-agent platforms
You sell a chatbot platform
Your customers already keep their answers, documents and integrations in your product, and the phone is the channel they ask about next. You offer it as part of your platform, under your brand and at your price, and every customer gets its own organisation on our side.
agencies and integrators
You build chatbots for clients
You know your clients' knowledge and processes, and you built the bot that holds them. The same knowledge can answer their phone, as a service you deliver and invoice yourself. How we work with agencies is described on its own page.
Why now
Buyers increasingly expect the phone from the platform they already use
More and more customer-service and chatbot platforms offer voice to their own customers. A business that runs its support on your platform will ask you the same question.
Voice in customer-service platforms
of Dutch consumers prefer the phone for customer service. For a chatbot, that figure is 0%.
Nationale Voice Monitor 2026
choose the phone when their question needs advice.
Nationale Voice Monitor 2026
of Dutch companies with 50 to 250 employees used AI in 2025, up from 20% in 2023. At 250 employees or more it is 66%.
CBS, December 2025
The Nationale Voice Monitor 2026 surveyed more than 1,000 Dutch consumers and was produced by a market research agency together with an AI consultancy. All sources are listed at the bottom of this page.
How it connects
The assistant holds the conversation. Your chatbot supplies the answer.
A phone call needs things a chat window does not: short spoken answers, turn-taking, interruptions, a transfer to a person. The voice assistant on VoiceDock takes care of those. When the caller asks something your chatbot knows, the assistant asks your chatbot during the call and puts the answer into words for the phone.
Caller
on a phone number or a talk button
VoiceDock assistant
listens, speaks, transfers
Your chatbot
knowledge, documents, your customer's systems
The voice assistant handles
- Listening, turn-taking and a caller who interrupts
- Answers short enough to say, rather than written to be read
- A short line while the lookup runs, so the caller does not hear silence
- A transfer to a person, warm or cold, when the bot cannot help
- Telling the caller they are speaking to an AI assistant
Your platform keeps
- The content: knowledge base, documents, the answers and their tone
- Updates in one place, for the chat and the phone alike
- The connections to your customer's own systems
- Your customer's account, settings and data
A chat answer is written to be read
Two ways to connect
A webhook tool or your own MCP server
Both are standard parts of the platform. Which one fits depends on what your platform already offers. We build the connection with your team once, then set it up per customer.
| Webhook tool | MCP server | |
|---|---|---|
| How it works | A tool in the assistant's configuration that points at a URL on your side. During the call the platform sends the question and the call context there, and your endpoint replies with the answer. | Your platform's MCP server, registered once with its address and headers, and attached to an assistant with the tools it may use. |
| What you build | One HTTP endpoint that takes a question and returns an answer. | Nothing new if your platform already runs an MCP server. Otherwise a thin MCP server on top of your API. |
| Authentication | Requests can be signed with a webhook secret, so your endpoint can check they come from us. | The headers you set, such as an API key, stored encrypted and never shown again. There is no sign-in flow to set up. |
| Per customer | One endpoint can serve every customer: the request carries the call context, such as the number that was called, so your platform knows which customer's bot to ask. | A server per customer organisation with that customer's own key, or one server with the key passed for each call. |
| If your platform is unreachable | The assistant is told the lookup failed, so it can say so and offer a transfer instead of guessing. | When the server is marked as required, the call is refused before it is answered, with the reason in the call log. Otherwise the call goes ahead without its tools. |
| Documentation | Custom tools and the tool-call webhook | MCP servers |
Webhook tool
- How it works
- A tool in the assistant's configuration that points at a URL on your side. During the call the platform sends the question and the call context there, and your endpoint replies with the answer.
- What you build
- One HTTP endpoint that takes a question and returns an answer.
- Authentication
- Requests can be signed with a webhook secret, so your endpoint can check they come from us.
- Per customer
- One endpoint can serve every customer: the request carries the call context, such as the number that was called, so your platform knows which customer's bot to ask.
- If your platform is unreachable
- The assistant is told the lookup failed, so it can say so and offer a transfer instead of guessing.
- Documentation
- Custom tools and the tool-call webhook
MCP server
- How it works
- Your platform's MCP server, registered once with its address and headers, and attached to an assistant with the tools it may use.
- What you build
- Nothing new if your platform already runs an MCP server. Otherwise a thin MCP server on top of your API.
- Authentication
- The headers you set, such as an API key, stored encrypted and never shown again. There is no sign-in flow to set up.
- Per customer
- A server per customer organisation with that customer's own key, or one server with the key passed for each call.
- If your platform is unreachable
- When the server is marked as required, the call is refused before it is answered, with the reason in the call log. Otherwise the call goes ahead without its tools.
- Documentation
- MCP servers
On either route, the assistant only uses the tools you list in its configuration.
A tool call on a live phone line has a caller waiting on the other end, which changes how you design it. Why it gets one attempt, and what the assistant says while it waits, is written up separately.
What your customers get
The phone, from the same knowledge as the chat
Four things you can offer, on the same assistant and the same connection to your platform. Most partners start with the first one.
A number that answers
The assistant answers your customer's number and handles questions from the same knowledge as the chatbot. That can be the customer's own number over their own SIP trunk, or a number we arrange on request.
A talk button next to the chat
A voice conversation in the browser, on the page where the chat already sits. It runs on the same assistant, at the same € 0,07 per minute of orchestration as a phone call.
A person when the bot cannot help
The assistant transfers the caller, warm or cold, to the customer's own team or to another number. Where transfers go wrong, and how to test for it, is written up separately.
Calls going out
The assistant can also make the call: a callback a visitor asked for in the chat, started from your platform through the API, with the same knowledge behind it.
The commercial model
You set the price, we invoice you for the platform
Each of your customers is an organisation of its own, with separate assistants, numbers, calls and usage. Your parent organisation covers their usage, with a price per organisation, so you decide what each customer pays. Our fee is € 0,07 per minute of orchestration for inbound, outbound and web calls. Model usage runs at cost on our keys, or on your own provider keys at the same € 0,07. Telephony is billed per leg at the carrier's rate. The work we do with your team, such as the connection to your platform and the first assistants, is billed separately and scoped with you up front. What a minute costs in total is worked out in full, and the pricing page lists every line.
Who does what
You keep the customer and the content, we run the voice layer
The split we start from. The details, including how support works, are agreed with you before the first customer goes live.
You
- The customer relationship, the offer and the invoice
- The chatbot, its content and its connections
- First-line contact with your customer
- Your price per customer
We
- The orchestration layer, and keeping it running
- Building the first assistants with your team, in working sessions
- Numbers, SIP trunks and transfers, and reading the SIP logs when a call goes wrong
- Failover between data centres, drilled on live traffic
Together
- The connection between the assistant and your platform
- The first pilot customer and what it has to prove
- How callers hear they are speaking to an AI, which the EU AI Act requires
- Who your customers call when something is off
A pilot
Start with one customer and one number
We start small on purpose, so the first customer proves the connection before the next ones follow.
- 01
Partner call
Thirty minutes with the engineers who build and run the voice layer. We look at your platform, its API, and how you want to sell the phone channel. - 02
The connection
We choose with you between a webhook tool and your MCP server, and build the connection to your platform together. It is built once and used for every customer after that. - 03
One pilot customer
You pick a customer whose chatbot already answers well. We build the assistant with you: the greeting, what it asks your chatbot, when it transfers, and how it tells callers it is an AI. Test calls come first, real callers after. - 04
Live, then the next customer
The number goes live. Each following customer gets its own organisation on the same connection, which you create yourself, by hand or through the API in your own onboarding.
Questions
What chatbot platforms and agencies ask
How does our chatbot connect to the voice assistant?
Through a tool the assistant calls during the conversation. That is either a webhook tool, an HTTP endpoint on your side that receives the question and returns the answer, or your own MCP server, registered once and attached to the assistant with the tools it may use. Both are documented, under custom tools and MCP servers. The connection to your platform is something we build with your team, not a switch you flip.
Does the content stay in our platform?
Yes. The assistant does not take a copy of your content. It asks your platform during the call, for the question the caller just asked, and you keep managing the content where it already lives. What is said on the call ends up in that call's transcript, like any call on the platform.
Can we sell it under our own brand, at our own price?
Yes. Your customers buy it from you, and each one is an organisation of its own under your parent organisation, with your price per organisation. A full white-label setup, including your own branding, is arranged with us.
Which phone numbers can our customers use?
Their own number over their own SIP trunk, which we set up with you, or a number we arrange on request. The assistant can transfer a call, warm or cold, to a person or another number, and it can make outbound calls. Telecom and VoIP providers have a page of their own.
Where is the data processed, and is there a DPA?
Orchestration, dashboard, API and storage run on EU infrastructure. You choose the model providers, and the Trust Center lists where each one processes data. A standard data processing agreement under GDPR Article 28 is ready to sign at no additional cost.
What does it cost?
€ 0,07 per minute of orchestration for inbound, outbound and web calls, plus model usage at cost on our keys, or billed directly by your provider on your own keys. Telephony is billed per leg at the carrier's rate. There is no per-seat licence, and you set the price your customers pay. The work we do with your team, such as the connection to your platform and the first assistants, is billed separately and scoped with you up front. Usage details are on the pricing page.
How does a pilot with one customer work?
You pick one customer and one number. We build the connection to your platform and the assistant with your team, test with calls before real callers are routed to it, and go live when you and your customer are satisfied. The build is scoped with you up front, so you know what the pilot involves and costs before it starts.
Does the assistant tell callers it is an AI?
Yes. We set up the greeting with you so callers hear they are speaking to an AI assistant, which Article 50 of the EU AI Act requires since 2 August 2026. What the article asks of voice agents is written up separately.
Bring us your platform and one customer
Thirty minutes with the engineers who build and run the voice layer. We look at your platform, your API and your customers, and tell you honestly whether the phone channel fits and what the first version should be.