Call Summary Agent · Use case
How can we use an AI agent to draft the after-call note from a call transcript on our own infrastructure?
A call summary agent reads the transcript of one call and drafts the after-call note in your format: the reason for the call, what the adviser and the customer agreed, and what is still open, each with the passage of the transcript behind it. The adviser who took the call checks the draft, corrects it and saves the note. The agent saves nothing by itself, changes no ticket, says nothing about how the adviser took the call, and reads no mood from the voice or the words.
The call ends.
The note is typed from memory.
- ✗ 01After each call the adviser types up what was said while the next caller waits, and the note is the only record the next adviser will have.
- ✗ 02A promise made on a call and missing from the note is a promise the company does not know it made.
Where it stops for a person
Always. The adviser checks and saves the note.Always. It records what was said, it does not judge it: the note says nothing about how the adviser took the call, gives no score and no rating, and describes no one's mood or tone. A card number or other payment detail spoken on the call is left out of the note. The adviser, who was on the call, is the check on the draft.
It drafts the note.
The adviser checks and saves it.
Step · 01
A call ends and its transcript is uploaded, or read through your own APIs.
Step · 02
The agent reads the transcript and your note format.
Step · 03
It drafts the note: the reason for the call, what was agreed, what the company promised and by when, and what is open.
Step · 04
It gives the passage of the transcript behind each line, and marks anything it could not make out.
Step · 05
The adviser who took the call checks the draft against what was said, corrects it and saves the note in your own system.
Reads from · connectors
The transcript of the one call, as an uploaded file or read through your own APIs with the operations that change something left out; and your note format and code list, as a knowledge source. It reads the transcript, not the audio: turning speech into text is done before it, by the speech recognition you run or choose.
Acts through · MCP servers
Nothing. It drafts a note. It has no tool that saves to a ticket, changes a record or sends anything to the customer.
What you set up
- Give the agent your note format and the code list, as a knowledge source.
- Decide how transcripts reach it: uploaded by the adviser, or read through the API of the system that holds them, with only the operations that read.
- Build the agent in Katonic Studio and test it on invented transcripts, with advisers comparing each draft against the call.
- Tell advisers and, where you have them, staff representatives that notes are drafted this way, and publish it to one team first.
Where it runs
The agent, its knowledge store and the model run on infrastructure you own or control. You choose the model and serve it yourself, so questions, documents and answers are not sent to a public AI service. Transcripts and notes stay inside the company. Where the speech recognition that makes the transcript is a hosted service, the call audio goes to that service: that is your choice of service, made before this agent runs.
How you get it today
This page is labelled Use case: the agent is built in Katonic Studio. You describe the job, connect what is named above, add test questions and release the agent once it passes them. When the Catalog holds a tested template for this job, this page will carry the Template label and show how to take it.
Does it save the note by itself?
No. It drafts. The adviser who took the call checks the draft and saves the note.
Is the summary used to assess the adviser?
No. The note says what the customer asked and what was agreed. It gives no score and says nothing about how the call was taken.
Does it listen to the call?
No. It reads the transcript. Speech recognition is a separate step that you run yourself or choose, and a hosted speech service receives the call audio.