Why file storage needs rethinking for AI work

Most of the files a team works on live in a cloud drive. Google Drive, Dropbox, OneDrive and SharePoint share a model that has worked for twenty years: one person opens a file, changes it and saves it. The drive keeps the latest copy and, if you're lucky, a list of older versions of that one file.

That model breaks as soon as an AI agent joins the team.

Agents don't edit one file at a time

Ask an agent to update a client proposal and it might touch the proposal, the budget spreadsheet, three supporting notes and the project plan in under a minute. The drive sees six unrelated saves. Nothing ties them together as one change you can look at, approve or undo.

Now put two people on the same folder, each with their own agent. Both start from Monday's files. Both save. One overwrites the other, or the drive gives up and creates budget (conflicted copy).xlsx. Someone has to open both and work out which numbers are right.

And when a run goes wrong, it goes wrong everywhere at once. Some drives can roll a whole folder back to a point in time, but that also throws away everything your teammates saved since. There's no way to undo just what one agent run did.

Once agents do real work on shared files, these stop being edge cases.

What storage has to do now

Storage for AI work needs a few things cloud drives weren't designed around:

1. A change is a set of files. When an agent edits twelve files, that's one change with one description, and you can see exactly what's in it.
2. Versions cover the whole workspace. Each change records the state of every file, so you can go back to the workspace as it was before any change.
3. Bigger work happens on a copy. An agent should be able to try something without touching the version everyone relies on. You look at the result and decide whether it goes in.
4. Conflicts are shown to you. If two people changed the same file, you see both and choose. Nobody's work disappears quietly.
5. Agents can find things without downloading everything. Search has to reach inside documents and spreadsheets, because that's where the answers are.
6. Agents get the same access as the person they work for. An agent working for a contractor should see the folder the contractor works on and nothing else.

Software teams solved most of this years ago with Git. But Git expects you to know Git, and it struggles with the large spreadsheets, PDFs and decks most teams work with. The ideas are right. The tool was built for programmers.

How Wiele works

Wiele is shared cloud storage built on those ideas, for teams whose AI agents work on real files.

• Your files stay ordinary files. Download a whole folder or a single file to your computer, or to the sandbox your agent runs in. Edit with Excel, Word, a text editor or any agent. There's no special app and no mounted drive.
• Saving is one step. When a change is ready, you or your agent push it. Every file in it becomes part of one new version of the workspace, with a note saying what changed.
• Every version you push is kept. Any file, or the whole workspace, can be brought back from an earlier version. Big files are fine too: a single file can be up to 10 GB.
• Copies for bigger changes. Anyone who can read a workspace can make a copy, let an agent work there and propose it back. Someone with access reviews what changed and merges it. Quick fixes can still go straight in if you're allowed to make them.
• Conflicts come to you. If two changes touch the same file, Wiele stops and shows you both. It never picks a winner for you.
• Staying in sync. Keep a folder up to date automatically, or let an agent's edits upload on their own while it works on a copy.
• Search inside your files. Keyword and meaning-based search across Markdown, Word documents, spreadsheets, CSVs and PDFs. Each result gives the file, the version and where in the file the match is. You don't download anything first.
• Permissions down to a single file. Share a workspace, a folder or one file, with read, write or admin access.
• Works with your agents. Wiele is a command-line tool with ready-made agent skills and an MCP server. Claude Code, Codex, Cursor or your own agents can use it, and your agent can install and set it up for you. It runs on macOS and Linux, including the Linux sandboxes many agents run in.

There are things Wiele deliberately doesn't do. It isn't a web editor, and it won't merge two edited Word documents by guessing. When two people change the same document, you decide which version wins or combine them yourself.

Where we are

Wiele is in early access. We're letting teams in a few at a time, starting with teams whose agents already work on shared documents and spreadsheets every day. If that sounds like yours, join the waitlist and we'll email you when your spot opens.