Reconciliation Break Agent · Use case
How can a bank use an AI agent to investigate reconciliation breaks on its own infrastructure?
A reconciliation break agent takes an item that did not match. It gathers both sides of the item and proposes the match or the adjusting entry, with the evidence. Nothing is posted without a person's approval.
The matching engine clears most items.
People clear the rest.
- ✗ 01The items an engine leaves are the ones where both sides have to be read: a different reference, an amount split in two, a difference in timing.
- ✗ 02A break nobody has explained carries forward to the next day, and ledger data cannot be sent to a public AI service.
Where it stops for a person
Always, before a match or an entry is posted.A reconciler approves or rejects each proposed match or adjusting entry. Nothing is posted without approval.
It gathers both sides.
A person posts.
Step · 01
A reconciler hands the agent an unmatched item.
Step · 02
The agent reads the item and looks for its other side, by amount, date, reference and counterparty.
Step · 03
It sets out what it found on each side and how the two relate.
Step · 04
It proposes the match, or the adjusting entry, with the evidence for it.
Step · 05
The reconciler approves or rejects. Nothing is posted without approval.
Reads from · connectors
Your own APIs
Your reconciliation system, through its own API. Or the database behind it, through a database connection or an MCP server that runs in your own cluster.
Acts through · your own APIs
The reconciliation system's posting operation, as a tool held for approval.
What you set up
- Import the reconciliation system's API description, or connect its database with a user that can only read.
- Keep posting held for approval.
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. Ledger data stays inside the bank.
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 the agent post entries?
No. It proposes. A reconciler approves each match or entry before anything is posted.
Does it replace our matching engine?
No. The engine keeps doing the volume. The agent works on what the engine leaves.
Can it read the ledger database directly?
Yes, to read. Connect the database with a user that has read access only, so the database enforces that as well.