A great deal of real business work lives in files: the spreadsheet that tracks open claims, the SOP that says how to handle an exception, the vendor's PDF invoice, the Word template that produces the customer letter. A thunk is designed to work with those files directly, in the file systems where they already live — not to make you copy them somewhere else first.
This section covers everything the platform can do with files. Start here for the map, then follow the links for detail.
Where files come from
A thunk can reach files from three places, and all three can be used together in the same workflow.
Microsoft OneDrive and SharePoint. Connect a Microsoft 365 account and your agents can read and write files in personal OneDrive and in SharePoint document libraries.
Google Drive. Connect a Google account and your agents can read and write files in My Drive and in shared drives.
Your local drive. Every thunk accepts files uploaded straight from the computer you are sitting at. This needs no connection at all — it is always available. Uploaded files are stored with the thunk and can be used everywhere a file can be used.
On top of those three, an agent can use any file that is reachable by a URL. If you can paste a link to it, an agent can usually open it and read what is inside.
Where files show up in a thunk
In your data. A column of a work item — or an input or output of a tool — can be of type File, holding a link to a document. There is a related Image type for visual content, which displays as a picture rather than a link. A File column can declare what type of file it expects and whether agents are allowed to modify it. See The File and Image Property Types.
In your instructions. An AI instruction can link to a file in one of the thunk's content folders, so the agent reads that exact document while it works. See Using Files in Instructions and Content Folders.
In content folders. Content folders hold curated collections of reference material — policies, product documentation, past cases. The platform indexes what is inside those files and gives your agents search tools over them, so an agent can find the relevant passage in a thousand-page corpus.
What an agent can do with a file
Read it. Given a link, an agent can open and read PDFs, Word documents, Excel workbooks and CSVs, PowerPoint decks, web pages, emails, and images. Long files are paged rather than truncated, so an agent can work through a several-hundred-page contract or a very large spreadsheet.
Work with spreadsheets. Agents can search a sheet, read a window of rows, append rows, update rows, add and remove columns, and create new sheets and workbooks — in Google Sheets and in Excel on Microsoft 365. See Working with Spreadsheets.
Work with documents. Agents can create Word documents and Google Docs, append content to them, export them to PDF, read long PDFs page by page, and fill in PDF forms. Formatted output — offer letters, confirmations, notices — is usually produced from a template. See Working with Documents and PDFs.
Work at the drive level. Agents are not limited to the contents of a file. They can create folders, move and copy files between them, upload files, share files, look up file details, and search a drive for the file they need. See Drive-Level File Operations.
A word about spreadsheets as a database
Spreadsheets are so convenient that it is tempting to run a whole business process out of one. The platform supports this deliberately: a Spreadsheet content folder turns each tab of a spreadsheet into database-style tools — list, look up, insert, update, delete — that agents can use without knowing anything about cells or ranges.
It is an excellent way to prototype. It is not a sound long-term store for enterprise data. Spreadsheet Content Folders explains both the capability and where its limits are.