Common requests are known.
Questions, approved answers and required details can be derived from real calls.
All services
A specific AI production use case
A phone assistant that answers. And knows when a person should take over.
The assistant answers inbound calls, handles repeat questions from your approved content, captures the request and contact details, and passes everything else to your team as structured data. No black box: answers, costs and handovers remain traceable.
German & EnglishAvailable 24/7One-off 3,900 €
What you get
- A production assistant for inbound calls
- Answers from your approved knowledge base
- Qualification and handover into one target system
- Measured cost, transcripts and alerts on failures
Scope & working together
A clearly bounded fixed-price package: live within four weeks from the day the number, content and target system are ready.
Remote from Germany. Straight with me, no agency in between.
The starting point
A missed call often means a lost request.
Many calls concern the same things: opening hours, status, responsibility or routing a request. They take time even though the answer is already documented. At the same time, an assistant must not guess when its knowledge base has no answer.
That takes more than a voice attached to a model. The important parts are a bounded knowledge base, clear escalation to a person, structured handover into existing workflows and operations where silent failures are noticed.
Typical starting point- Calls are lost outside business hours.
- The same questions are answered repeatedly every day.
- Call-back notes contain too little context.
- Information is copied into a CRM or ticketing system by hand.
- Nobody knows what an existing voice bot costs per conversation.
- Nobody notices immediately when the assistant stops answering.
Who this is for
For teams where recurring calls create work and availability matters.
For service, sales and operations owners with regular inbound call volume. Many requests repeat, information exists and a clean handoff to people matters more than endless artificial conversation.
Missed calls have consequences.
Leads, appointments or service cases are lost or create callback queues for the team.
A person remains available.
Limits, escalation and handoff are part of the process; the assistant supports the team.
What I do
Answer.
Understand.
Hand over.
01
Conversation scope and boundaries
We define which requests the assistant handles, what it captures and when it hands over to a person. What it does not know, it does not invent.
02
Answers from your content
Answers come from an approved knowledge base rather than the model’s general knowledge. Content can be updated without rebuilding the whole assistant.
03
Qualification and handover
The request, urgency and contact details are captured as structured data and passed to CRM, ticketing or email with a transcript.
04
Telephony and operations
The number, language model and target system are connected. Cost per call is measured, failures trigger alerts and conversations follow the agreed data-protection framework.
Which calls should the assistant handle?
In 30 minutes we establish whether the use case fits the package.
How it runs
A production assistant in four weeks.
The scope stays clear: one assistant, one target system and defined conversation paths.
STEP 01
Scope the requests
Repeat questions, handover rules, language and target system are agreed.
STEP 02
Build knowledge and dialogue
Approved content, conversation flow and boundaries are configured and tested.
STEP 03
Connect the systems
Telephony, target system, transcripts and cost measurement are connected.
STEP 04
Test and go live
Tests with realistic conversations, a walkthrough and a controlled production launch.
Entry offer
AI phone assistant. Fixed price 3,900 €
one-off, no monthly cost from me
The assistant answers calls, handles what repeats, qualifies the rest and files it where your team already works.
What you get
- Inbound calls answered in German and English
- Answers from your own approved content
- The request, urgency and contact details captured
- Handover into one target system with a transcript
- Escalation to a person under rules you define
- Cost measurement per call and alerts on silent failures
- Configuration, prompts, knowledge base and integration in your accounts
What you do not get
- No outbound sales calls
- No more than one target system in the fixed price
- No languages beyond German and English
- No migration of existing data from a current phone system
- A one-off payment: no subscription, licence or monthly invoice from me
- Four weeks from the day the number, content and target system are ready
- Language-model and telephony usage is billed to you directly by the providers
- Data processing and caller information are settled before the first call
- The assistant and its configuration belong to you
The outcome
What changes afterwards.
Available without a queue
Repeat enquiries are accepted and answered around the clock.
Handovers with context
Your team receives the request, urgency and conversation rather than only a number to call back.
A system you own
Configuration, prompts, knowledge base and integration sit with you rather than a reseller’s account.
Technologies I use
The building blocks.
Voice
- Speech-to-text
- Text-to-speech
- German
- English
AI
- Amazon Bedrock
- Claude
- Gemini
- RAG
- Guardrails
Integration
- REST
- Webhooks
- CRM
- Ticketing
- Slack
Operations
- CloudWatch
- Transcripts
- Alerts
- Cost per call
Running it myself
It answers my own business number.
The assistant I offer here is one I run myself. Call +49 (2451) 6123001 and fail to reach me, and you do not land in voicemail. The assistant picks up, says at the start that you are speaking to an AI system, answers questions about my work from the content of this website, takes down what you need and your contact details, and offers an appointment if you want one. It has worked that way since June 2026.
It is the fallback, not the front line: if I can pick up, I pick up. That split is the right one for me, because a call here is usually meant to become a conversation. Where the same twenty questions arrive every day the line sits somewhere else, and where it sits in your case is something we decide together.
My own version runs on Twilio, a server at Hetzner and two language models from Google and Mistral, with all processing inside the EU. Yours would look different if everything already sits on AWS. There is deliberately no metric here: I have none I could substantiate. You can still check the claim, by calling.
Who you are talking to
Directly with me as a freelancer. No agency in between.
I am Tim Rutte. More than 20 years in software development. Today I take AI systems into controlled production. You talk to the person who touches your code, from the first call to the handover.
- 20+years in software development
- 50+successful projects
- 2003working remotely since then
Common questions
Questions about the AI phone assistant.
Does the assistant replace our staff?
No. It handles repeat requests and collects context. As soon as a conversation falls outside the agreed scope, it hands over to a person.
Can it give a wrong answer?
A bounded knowledge base, clear answer rules and handover on uncertainty reduce the risk. No language model can guarantee zero errors, which is why boundaries and transcripts are part of the system.
Can it use our existing number?
That depends on your phone system. Cloud telephony can usually connect directly; with an on-premise system we establish whether forwarding or a new number makes more sense.
What are the running costs?
The language model and telephony are billed by usage directly by their providers. Cost per call is measured and made transparent before launch.
Where are calls and transcripts stored?
On request, entirely inside your infrastructure and accounts. Location, retention and access rights are agreed before the first production call.
Other services
What else I help with.
AI in production
The infrastructure behind AI systems that actually ship: MCP servers, controlled tool access, LLM integration with real permissions and cost control.
Learn moreProcess automation
Replacing manual workflows with real systems: wired into the software you already run, with permissions and an audit log instead of a chain of tools.
Learn moreAutomating invoice processing
From the incoming document to the posted entry without retyping: read XRechnung and ZUGFeRD, match against the purchase order, route the approval, hand over to your ERP, with a full trail.
Learn more