Your files & how your assistant remembers

Files are just files

The files and folders in a workspace work exactly like the ones on your computer — documents, spreadsheets, images, and everything else. You can open them in the viewer, rename them, move them between folders, download them, and share them with other people.

There's nothing special to learn here: if you know how files work on your laptop, you already know how they work in your workspace. (For opening and previewing files, see The app; for sending a file or folder to someone else, see Sharing.)

Your workspace is the assistant's memory

The whole workspace is your assistant's knowledge base. What it saves as files it can build on later; things mentioned only in passing in a chat it won't reliably remember.

So when you tell it something worth keeping — a decision you've made, a preference about how you like things done, a plan you've agreed on — it files that away as a document rather than leaving it buried in the conversation. A note written down in the workspace is something the assistant can come back to weeks later; a remark dropped into a chat may quietly fade.

This is why a well-kept workspace makes the assistant smarter over time. The more of what matters lives in files, the more your assistant has to draw on.

A few files you may notice

As your assistant works, you might spot a few files it keeps for itself. You never have to write or maintain these — the assistant creates them and keeps them up to date automatically. They're worth knowing about, though, because they explain how it stays organized.

• AGENTS.md — the standing instructions that tell the assistant how to organize this workspace: its house rules. You're welcome to read it, and to tweak it if you'd like to change how things are arranged or styled.
• index.md — a hand-maintained map of the workspace, so both you and the assistant can find things quickly.
• log.md — a running history of what changed and why.
• knowledge/memory.md — the rules you've taught it. See below.

If you'd rather never think about these, you don't have to. They quietly keep the workspace tidy in the background.

What it remembers, and how to correct it

When you tell your assistant how you want something done — "never post the documents, always leave them as drafts", "invoices go in the client's folder", "reply in Russian" — it writes that down straight away and says in one line that it did. From then on it applies the rule without being reminded, in this workspace, for everyone who works here.

You don't need any special wording. Just correct it the way you would a colleague, and carry on.

Everything it has learned lives in one short file, knowledge/memory.md, and you can open and edit it like any other file. If it has written down something that isn't quite right, change the line or delete it — your edit wins. If it picked up a rule you never meant to make permanent, that's how you take it back.

Two things worth knowing:

• The memory is deliberately short, so it stays reliable. When it fills up, your assistant tidies it rather than dropping anything: it merges related rules, and when a rule has grown a tail of detail — ids, addresses, a list of steps — it moves that detail into a knowledge/ note and keeps the short rule pointing at it. It tells you what it moved and where.
• A new rule takes effect from the next conversation. The assistant already knows it for the rest of the current one, so you won't have to repeat yourself either way.

Getting files in (uploading)

There are a few ways to bring your own files into a workspace:

• Drag and drop files onto the sidebar — a "Drop to upload" area appears as you drag.
• Or use a folder's menu: "Upload Files" and "Upload Folder".

Files up to 2 GB each are supported this way, and you'll see a live progress list at the bottom of the sidebar while they upload. Files you upload this way land in your workspace, where the assistant can work on them as part of your ongoing work.

You can also attach files to a single chat message — the paperclip / "Attach file" option — when you just want the assistant to look at something for that one request. (See Chats for the size limits on attachments.)

How a well-kept space tends to look

There's no required layout — your assistant adapts to how you work. But over time, spaces tend to settle into a shape that's easy to navigate. As a rough illustration, a tidy workspace often grows:

• a folder per topic or per project, so each body of work stays self-contained;
• a folder per person for family or team spaces;
• a decisions/ or knowledge/ folder for the things worth remembering;
• an uploads/ folder for files you've added yourself;
• a scripts/ folder for connected integrations (see Workspace scripts and Integrations).

Finished reports are often saved as HTML files you can open right in the viewer — see Deliverables. And when the assistant keeps structured, persistent data for you (the kind that powers a little dashboard), that's covered in App data.

This is just how spaces tend to settle, not a template you have to follow. Keep things however suits you, and the assistant will follow your lead.