Files & folders agents can read

A projectis a folder on your machine, and agents can work with what’s inside it — not just the document you have open. That’s what lets them write from your own material instead of guessing.

Everything in your project

Whatever you keep in the project folder is available to agents as context: your draft, the sources in your library, your notes, and your data. When an agent drafts, checks a citation, or answers a question, it can draw on all of it — so its work is grounded in your project, not generic knowledge.

Sources, data, and results

This includes folders of results and outputs. Drop a folder of result files into the project — exported data, generated figures, tables, a batch of PDFs — and agents can read across it. For example, you can ask an agent to:

• summarize what a folder of result files shows;
• pull specific numbers from your data into your draft;
• compare two sets of results and describe the difference;
• turn a results table into a figure or a paragraph of prose.

Keeping your results inside the project folder is all the setup needed — once they’re there, they’re readable.

Pointing an agent at files

You can let an agent find what it needs, or point it precisely:

• @-mention a specific file or folder in chat to focus the agent on exactly that.
• Or just describe it— “summarize the results in the results folder” — and the agent reads what’s relevant.
• Use Super Search to find a passage across everything in the project, then act on it.

What agents can read

Agents read text documents, PDFs, Office files, spreadsheets and data files, images, and more — see file format support for the full list. Reading happens on your machine; your files stay local and are only sent to a model when a request needs one. See privacy & data.