An AI agent for Google Sheets: update and question your sheets by sending a message

What an agent can do with a sheet

Once your Google account is connected, the agent can work with any spreadsheet that account can open. These are the jobs that come up most, each one something you can ask for in a sentence.

Add a row from a message

You type “Sold 3 of the large planters to Mrs Okafor, 45 each, paid cash” and the agent adds a row to your orders sheet with the date, customer, item, quantity and amount. It can append to the end of a table, so you do not have to say which row.

Log an expense from a photo

Send a photo of a receipt in Telegram and ask it to log the date, shop and total in your expenses tab. This relies on the AI model being able to read images, so try it on a few receipts before you trust it, and check the totals it reads.

Look up a row

“When did Mrs Okafor last order, and what was it?” The agent can query the rows for that customer and sort them by date to find the latest. A quick lookup only returns the first row whose cell matches exactly, so “Okafor” will not find “Mrs Okafor”; it helps to say how the name is written in your sheet.

Total or count a column

“How much did I spend on materials this month?” The agent can find the rows where one column matches a value, such as the category, and add up another column. It can also run a query in SQL-style language against a table in the sheet, which suits questions like “top five customers by value”.

Build a summary on a schedule

Ask for a job such as “every Monday at 8:00, read last week’s rows, total sales and expenses, and message me the result”. Hermes can run jobs on a schedule and send the output back to the chat where you set it up. It can also draw a chart in the sheet from a range of your data.

Find a file, or start a new one

It can search your spreadsheets by name or content, which helps when you can’t remember which of five “Orders” files is the live one. It can also create a new spreadsheet from a list you paste in, using the first line as the column headings, or add a new tab to an existing one.

What it looks like in a chat

Take Ruth, who sells handmade candles at markets and online. She keeps one spreadsheet with an Orders tab and an Expenses tab, and she is rarely at a computer. The names and figures here are made up to show the shape of it.

• Ruth, from the market stall: “Add an order: Dana Reyes, 6 soy candles, 18 each, paid by card.” Agent: “Added to Orders, row 214: today’s date, Dana Reyes, soy candle, 6, 18, 108, card.”
• Ruth, on the train: “What did Dana order last time?” Agent: “Her previous order was row 187: 2 lavender candles on 14 September.”
• Ruth, photographing a wax delivery receipt: “Put this in Expenses under materials.” Agent: “Read it as 64.20 from the supplier on 29 October. Added to Expenses, row 91. Tell me if the total is wrong.”

The standing job she set up once, so the weekly summary arrives without her asking:

Every Monday at 8:00, read last week's rows from Orders and Expenses
in my Candles sheet. Message me total sales, total expenses and
the three biggest orders. Do not change anything in the sheet.

That last sentence matters. A summary job only needs to read, so say so. Keep the agent’s write jobs to the ones where you have told it exactly which tab and which columns to use.

Chat agent or in-sheet AI: which fits

They are built for different moments, and plenty of people use both. Google’s own help page for Gemini in Google Sheets (read 31 October 2026) lists what it does inside an open spreadsheet: creating tables and formulas, generating analysis and insights, building charts, and actions such as conditional formatting, pivot tables, dropdowns, filters, find and replace and inserting or deleting rows and columns. The same page says the feature requires an eligible Google Workspace or Google AI plan, so check what your own account includes.

A rule of thumb: if the job is “help me build this spreadsheet”, use the tool inside the sheet. If the job is “keep this spreadsheet fed and tell me what it says”, a chat agent suits better. A business that does both often keeps the structure and formulas built in the sheet, then lets the agent append to it. For the wider picture of what an assistant takes off an owner, see an AI assistant for small business owners.

Checking its edits and keeping a way back

An agent will sometimes put a value in the wrong column or misread a figure. Plan for that before it writes anything you rely on.

• Ask it to say what it changed. Have it reply with the tab, the row and the values it wrote, as in the chat above. A reply you can read in ten seconds is a cheap check.
• Start in a copy. Make a copy of the real spreadsheet, point the agent at the copy for the first week, and compare. Move to the real one when its rows keep coming out right.
• Use Google’s version history. Google’s help page (read 31 October 2026) says you open it from Last edit at the top right of the file, pick an earlier version and choose Restore this version. You need permission to edit the file to browse earlier versions. Open it after the agent’s first edits to see how its changes are shown.
• Name a version before a big job. The same page says you can name versions so they are not merged, up to 15 named versions per spreadsheet. It also warns that restoring an earlier version of a spreadsheet clears the cell-by-cell edit history for changes made before the restore, and suggests making a copy if you need that history.

The agent can also clear cell contents and delete rows, columns or whole tabs if you ask. Tell it in its standing instructions never to delete or clear anything without asking you first.

Limits and what to keep in your own hands

• It reaches what the connected account reaches. If you connect your main Google account, the agent can open every spreadsheet that account can. If that is more than you want, create a separate Google account, share only the relevant sheets with it, and connect that one.
• Test it on your workbook. Start with one tab or a limited range and check the results before handing it a large workbook. A complicated workbook, with many linked sheets and heavy formulas, is better edited by you in the sheet.
• Ambiguous instructions produce wrong rows. “Log the usual” will fail. Name the tab and the columns once, and keep your headings consistent.
• Money and records are yours to check. It can total a column, but a sheet you file accounts or tax from deserves a human look before you rely on it. The agent is not an accountant and does not give tax advice.
• Messages to other people stay under your control. If a sheet drives replies to customers, have the agent draft them for you to approve. Keep the messages to people who already contacted you or buy from you, and follow the rules of the messaging app you use.
• Photos and handwriting are hit and miss. Receipt reading depends on the AI model and on the picture. Blurry or crumpled receipts are the usual failures.

For messages that reach other people, or anything that triggers a payment, keep a person in the loop. If the sheet is where your customer email workflow starts, see can ChatGPT answer my emails?

Doing it with your own Hermes Agent

Hermes Agent is open source and includes a skill called Google Workspace. It covers Gmail, Calendar, Drive, Contacts, Sheets and Docs, and for Sheets it documents commands to create a spreadsheet, read a range, update a range and append rows. Its own instructions tell the agent to confirm with you before modifying Docs or Sheets, which is a sensible default for a first week.

The catch is setup. The skill uses your own Google Cloud OAuth client: you enable the Sheets, Docs, Drive and other APIs in a Google Cloud project, create an OAuth client, and authorise it once. That is comfortable for a developer and a lot for a shopkeeper. If you would rather start from a server you run, installing Hermes Agent is the first step, and MCP servers are the other route to a Sheets connection. You can then link Telegram with the Telegram guide, or WhatsApp with the WhatsApp guide.

Google Sheets on ReadyHermes

ReadyHermes runs a Hermes Agent for you, so you do not set up a machine or a Google Cloud project yourself. The Apps panel lists connected apps including Google Sheets, Gmail, Google Calendar and Google Docs. You press Connect, sign in to your Google account yourself, approve, and come back. You can disconnect the app from the same panel at any time. Then you ask in the chat, from the browser or from Telegram, WhatsApp, Discord, WeChat or Lark.

1. Start the free trial, which lasts 1 hour. A new assistant is ready in about 10 seconds.
2. Open Apps, find Google Sheets and connect it. Connect Telegram or WhatsApp so you can reach the agent from your phone.
3. Ask it to find your orders sheet, tell you its tabs and headings, and add one test row. Check the sheet.
4. Ask for the weekly summary job, with the instruction not to change anything.

When the trial ends the assistant pauses and is deleted unless you choose a plan within 24 hours. A paused assistant cannot connect apps, so choose a plan to carry on. ReadyHermes does not offer a spreadsheet add-on or a panel that appears inside your sheet; everything happens in the chat. The apps page shows the full list, and an AI agent for a one-person business shows how sheets fit into a whole working week.