Anthropic has added file editing to Claude, letting users work with Word, PowerPoint, Excel, and PDF documents directly in the chat.
You can now upload, create, and edit files right inside Claude's web or desktop app. The feature is available as a preview for Max, Team, and Enterprise customers. Pro users will get access in the next few weeks, according to Anthropic.
If you need a specific file, you can just tell Claude what you want and upload any data you need included. The model can generate ready-made documents like budget templates with working formulas, project trackers with dashboards, or financial models that feature scenario analysis. Anthropic says these files come with functioning formulas and can span multiple spreadsheets.
Claude can handle more complex jobs, too, like cleaning up messy data, running statistical analyses with charts, or adding explanatory text. It can also convert between formats: PDF reports can turn into slide decks, meeting notes can become formatted Word files, and invoices can be turned into structured Excel sheets with calculations.
All of this happens in a private computing environment where Claude writes and runs its own code to build the files, much like ChatGPT Agent.
Setup and security
To turn on file editing, users need to enable "Upgraded file creation and analysis" in the settings. After that, you can upload files or just describe what you want in chat. Finished documents can be downloaded or saved straight to Google Drive.
Anthropic recommends starting with simple tasks like data cleanup or short reports before moving on to more complex projects, such as financial models. It's worth remembering that these systems can and will make mistakes.
Claude works in a sandboxed environment with restricted internet access, where it can install software packages from public sources like npm or PyPI. This setup is meant to limit risk, but it isn't perfect. Anthropic cautions that manipulated files or websites could still trick Claude into running unsafe code or leaking sensitive information, such as data from prompts, projects, or connected services like Google Drive. These kinds of agentic risks are well documented.