Expenses logged to Google Sheets from a chat message

Expenses logged to Google Sheets from a chat message

An n8n workflow that logs an expense you type in n8n's chat. An agent calls a save step, a model reads the cost, the description and the date from your message, and a new row is added to a Google Sheet.

What is inside

• workflow.json: the workflow as n8n exports it, 10 nodes. Nine run, and one is a sticky note. Five of the nine belong to the chat flow: the chat trigger, the agent, its OpenAI chat model, its window memory and its save tool. The other four make the second flow: a Workflow Input Trigger, an information extractor with its own OpenAI chat model, and the Google Sheets step. It holds no credential.
• README.md: these notes.

What we changed from the original copy

• Both OpenAI Chat Model nodes had no model set, so they ran n8n's default for their version, gpt-5-mini. OpenAI's deprecation page, read on 12 November 2026, shuts that model down on 11 December 2026 and names gpt-5.6-terra.
• OpenAI Chat Model, under AI Agent, now uses gpt-5.6-terra. The agent calls a tool, and OpenAI's page for gpt-6-luna says Chat Completions supports function calling only with reasoning effort set to none, which n8n's chat model node never sends.
• OpenAI Chat Model1, under Expense text to JSON parser, now uses gpt-6-luna. That node puts the format instructions into the prompt and calls no tool.
• Parse msg and save to Sheets pointed at a saved workflow in the author's own n8n. Its Workflow field now holds the expression ={{ $workflow.id }}, which is this workflow, so the tool runs the second flow of the copy you import.
• The credential links are gone. The document and the tab in Save expense into Google Sheets are empty, so you pick your own. The sticky note is written again, without the link to the author's sheet.
• The file's name is ours.
• Nothing else. The agent's rules, the extractor's three fields and the sheet columns are as the file had them.

Mindket read this copy against n8n's source and OpenAI's pages. It was not run on a live n8n.

Before you start

• An n8n instance, on n8n Cloud or your own server.
• An OpenAI API key.
• A Google account that can edit your sheet.
• A Google Sheet with one tab that has a header row of date, cost, descr and msg.

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 and in OpenAI Chat Model1.
3. Add a Google Sheets credential and select it in Save expense into Google Sheets.
4. In that node, pick your document and your tab.
5. Check that Parse msg and save to Sheets points at this same workflow. The file sets its Workflow field to this workflow's own id.
6. Save and publish the workflow.
7. Open the chat, send "coffee beans 18.40 eur on 3 march" and check the sheet.

What happens in a run

1. When chat message received passes your message to AI Agent, which has Window Buffer Memory and OpenAI Chat Model.
2. The agent's system message makes it a helpful accountant. It must use the save tool for an expense message, and its reply is a fixed confirmation line followed by what the tool returned. It stops after 3 iterations.
3. Parse msg and save to Sheets is named save_expense_in_db for the agent. It runs the second flow of this workflow and hands it your message as input1.
4. Workflow Input Trigger starts that flow and passes input1 to Expense text to JSON parser.
5. Expense text to JSON parser asks its model to convert the expense to JSON with three fields: cost (a number, required), descr (the description, required) and date (a date in UTC format).
6. Save expense into Google Sheets appends a row with date, cost, descr and msg, the original message. When the model gives no date, the date is the time of the run.
7. The agent answers with the confirmation line.

What to change

• The agent's rules and reply: the system message in AI Agent.
• What the model reads from a message: the attributes in Expense text to JSON parser.
• The columns: Save expense into Google Sheets. If you add a field, add it to the extractor and the sheet.
• The models: the two OpenAI chat model nodes. Keep one that can call tools under AI Agent.
• The number of tool calls: the Max Iterations option in AI Agent.

Good to know

• The tool runs the saved version of the workflow, so save it after each change.
• A message with no date gets the time of the run as its date.
• The chat memory holds the last five turns inside the n8n process, and a restart clears it.
• Anyone who can open the chat can add rows to your sheet.
• Data: your messages go to OpenAI, and the rows go to your sheet.
• Cost: a few model calls per expense, billed by OpenAI. The agent calls its model before and after the tool, and the extractor makes one more.