Rent payments checked against bank CSVs and a local workbook

Rent payments checked against bank CSVs and a local workbook

An n8n workflow that watches a folder for bank statement CSV files. An AI agent checks each statement against tenants and
properties kept in a local Excel workbook. It flags missed or wrong rent, tenancies that end soon and unpaid final fees, and
it writes the alerts to a sheet of the same workbook. It needs a self-hosted n8n.

What is inside

• workflow.json: the workflow as n8n exports it, seventeen nodes: eleven that do the work and six sticky notes. It holds
  no credential.
• README.md: these notes.

What we changed from the original copy

• The credential link on OpenAI Chat Model is gone. Select your own.
• The model was gpt-4o. OpenAI's deprecations page, read on 12 November 2026, gives 14 January 2027 as the shutdown date of
  the gpt-4o snapshots and names gpt-6-luna. OpenAI's page for that model says Chat Completions supports function
  calling only with reasoning effort set to none, and this agent calls tools, so the node now names gpt-5.6-terra.
• The agent's instructions asked for a field named action required, with a space. The output parser and the last Code
  node use action_required, so the instructions now say that.
• The workbook name was spelled reconcilation-workbook.xlsx. It is reconciliation-workbook.xlsx now.
• The notes are rewritten as short notes about each step.
• Nothing else.

Before you start

• A self-hosted n8n. The trigger and the workbook code use the files of the machine that runs n8n.
• The Local File Trigger node enabled. n8n's 2.0 breaking changes, read on 12 November 2026, list it as disabled by
  default, and removing it from NODES_EXCLUDE brings it back.
• The xlsx module available to code. The Code node and both code tools call require('xlsx'). n8n's documentation says external
  modules need NODE_FUNCTION_ALLOW_EXTERNAL, and that the variable goes on the task
  runners when you use them.
• A folder that n8n can reach. The file uses /home/node/host_mount/reconciliation_project, a path in a Docker setup.
• A workbook named reconciliation-workbook.xlsx in that folder, with three sheets.
• tenants needs the columns Tenant ID and Tenant Name, plus the terms, rent amount and notes you want the agent to see.
• properties needs Property ID, plus the address, postcode and type.
• alerts needs a header row, and the workflow appends below it.
• Bank statement CSV files with the columns date, reference, money_in and money_out in the first row.
• An OpenAI API key.
• A backup of the workbook. The workflow rewrites it.

Set it up

1. In n8n, create a workflow, open its menu and choose Import from File. Pick workflow.json.
2. Add an OpenAI credential and select it in OpenAI Chat Model.
3. In Watch For Bank Statements, set the folder's path as n8n sees it.
4. In Set Variables, set the workbook's path.
5. Save, then press Publish.
6. Drop a statement CSV into the folder and wait for the run to finish.

How a run goes

1. Watch For Bank Statements fires when a file is added. Its ignore pattern, !**/*.csv, is meant to skip everything that
   is not a CSV.
2. Set Variables keeps the workbook's path.
3. Get Bank Statement File reads the CSV and Get CSV Data turns each row into an item.
4. Reconcile Rental Payments is an AI agent. It gets the statement as a table, and it can call Get Tenant Details and
   Get Property Details, two code tools that read the workbook.
5. The agent answers with a list of alerts, and Structured Output Parser holds it to one shape.
6. Alert Actions To List makes one item per alert, and Append To Spreadsheet adds one row each to the alerts sheet.

Change the look

• What counts as a problem: the system message of Reconcile Rental Payments.
• The model: OpenAI Chat Model.
• The watched folder: Watch For Bank Statements.
• Which columns the agent sees about a tenant or a property: the sheets themselves, since the tools return whole rows.

Good to know

• This workflow changes your workbook. Append To Spreadsheet saves a copy first, next to the workbook with .bak.xlsx added
  to its name. The same name is used each time, so the copy holds only the state before the latest run.
• The trigger does not wait for a file to finish writing, because its option for that is off by default. A large CSV may be
  read half written.
• The agent decides what is wrong. Check its alerts before you act on them.
• A long statement goes into the prompt in full and can pass the model's limits.
• The statement rows and the tenant and property records the agent asks for go to OpenAI.
• The JSON holds no credential.