telkit
Get started

Use case

AI phone agent in Norwegian: what telkit covers today

A phone agent that holds a conversation needs a number, audio both ways, speech to text while people talk, a language model and a voice. telkit has the number, the routing, the voice and the text after the call. The live conversation with a language model is not part of telkit.

Get startedAll use cases

Last reviewed 14 October 2026

The problem

Many people looking for an AI phone agent really want three things: an answer when nobody is free, the caller routed to the right person, and the call as text afterwards.

You can build that with telkit today. Your own system decides what happens to each incoming call, a Norwegian voice reads the message, and the caller is put through to a person. Afterwards the recording and the text are ready.

What telkit does not do is listen and answer freely within the same call. The API does not stream the audio to an agent you run yourself while the call is going on.

The building blocks telkit has

Voice and transcription, with the same key and balance.
  • Norwegian numbers and routingRent a number in Oslo, Bergen, Trondheim, Stavanger, Kristiansand or Tromsø. Choose what happens when someone calls: forward, a greeting then forward, a message, ring in the browser, or let your system decide.
  • Your system decidestelkit sends a signed request about the call to your system and waits 5 seconds. The answer is one of three: forward to a number, play an audio clip, or hang up.
  • A Norwegian voiceTurn text into audio with a Norwegian female or male voice, or place a call that reads a text aloud when someone answers.
  • Recording and text after the callRecord the call and get it written out when it ends. The recording and the text sit with the call in the dashboard.
  • Transcribing filesSend a file or an https link and get text with timestamps and who spoke, as JSON, SRT, VTT or plain text.

Example flows

Three set-ups you can build with what exists.

A switchboard after hours

  1. The number is set to «Til eget system» (your own system). When someone calls, your system gets a signed request saying who is calling.
  2. During opening hours your system answers forward to the mobile of whoever is on duty.
  3. After hours your system answers play with an audio clip saying when you open. When the clip ends, telkit hangs up.

An outbound message in a Norwegian voice

  1. Your system asks telkit to call the customer from one of the profile's numbers.
  2. When the customer answers, the text is read in a Norwegian voice, twice if you like. Then telkit hangs up.
  3. call.ended reaches your webhook with how long the call was connected and what it cost.

Text for your own language model

  1. The call is recorded, and the text is made when it ends.
  2. If you hold the audio file yourself, you send it to the transcription API and get text with timestamps and who spoke.
  3. Your system hands the text to the language model you use and stores the summary wherever you want it.

What the request looks like

What telkit sends your system when someone calls, the answer that puts the call through, and an outbound call in a Norwegian voice.
From telkit to your systemjson
{  "type": "call.incoming",  "id": "…",  "createdAt": "2026-10-14T21:14:00.000Z",  "data": {    "callId": "…",    "profileId": "…",    "numberId": "…",    "from": "+4791234567",    "to": "+4721000000"  }}
The request is signed like a webhook. If no usable answer arrives in time, telkit hangs up.
Your system's answerjson
{  "action": "forward",  "to": "+4799887766"}
The other answers are {"action":"play","clipId":"…"} and {"action":"reject"}.
POST /v1/callsbash
curl https://api.telkit.no/v1/calls \  -H "Authorization: Bearer $TELKIT_API_KEY" \  -H "Idempotency-Key: levering-2026-10-14-118" \  -H "Content-Type: application/json" \  -d '{    "profileId": "…",    "from": "+4721000000",    "to": "+4791234567",    "text": "Hei, dette er Rørleggeren AS. Vi kommer i morgen mellom klokka åtte og ti.",    "voice": "no_male",    "repeat": true  }'
The call is paid for before it is dialled, and it ends when the paid time is used up.

What to watch out for

A call with a machine comes with obligations, whether you build an agent or just a greeting.
  • Section 15 of the Norwegian Marketing Control Act requires prior consent for marketing to private individuals through automated calling systems. A message about an appointment the customer made is not an offer.
  • The Norwegian Data Protection Authority (Datatilsynet) writes that businesses as a rule must ask for consent before they record a call with a customer.
  • Say in the first sentence that the caller is hearing a machine. It is honest, and people speak more clearly when they know.
  • Decide how long recordings and text are kept before you start. They are personal data and must be deleted when their purpose is fulfilled.
  • Speech and transcription run through a subprocessor. See who, and where the data is processed.Subprocessors

What it costs

Calls are paid per started minute. Recording and text come on top, also per started minute of the call.
Phone number, per month
NOK 59
Incoming call, per minute
NOK 0.60
Outbound call, per minute
NOK 0.60
Recording, per minute of the call
NOK 0.20
Text of the call, per minute
NOK 0.12
Text to speech, per 1,000 characters
NOK 0.60
File transcription, per audio minute
NOK 0.06

An incoming 3-minute call with recording and text costs NOK 2.76 excl. VAT.

Prices exclude VAT.

See the full price list

Questions about AI phone agents

Can I buy a ready-made AI phone agent from telkit?

No. telkit has the building blocks around one: numbers, routing, a Norwegian voice, recording and text. A live conversation between the caller and a language model is not part of the service.

Can my system decide what happens to each call?

Yes. With the «Til eget system» routing your system gets a signed request for every call and answers within 5 seconds with forward, play or reject.

Which voices are there?

A Norwegian female voice, a Norwegian male voice and an English female voice.

Can a caller be put through to a person?

Yes. The call can be forwarded to a Norwegian number, or ring in the browser for the people logged in to the dashboard.

Where is the audio processed?

The audio passes through the subprocessors for telephony and speech. The subprocessors page shows who they are and where they process data.

Read more about the services

Other use cases

Set up your number

Register your company, order a Norwegian number and let your system decide what happens when someone calls.

Get startedSee pricing