We Love Claude

File Creation

Generate downloadable documents (PDF, Word, Excel, PowerPoint) directly from a chat.

Available on
Desktop Web

What it is

Claude can execute code in an isolated sandbox and hand you back real files — Excel spreadsheets with working formulas, PowerPoint decks, Word documents, PDFs. You prompt in natural language; Claude writes the code, runs it, and you download the result.

This is distinct from Artifacts. Artifacts are live web apps and documents that render in a side panel. File creation produces an actual .xlsx, .pptx, .docx, or .pdf you can open in Office.

Where to find it

On desktop

  • Settings → Capabilities → Code execution and file creation. Toggle on.
  • Then just ask in any chat: “build me a financial model”, “turn this into a slide deck”, “make a PDF report”.

On the web

  • Same setting path on claude.ai.
  • For Team and Enterprise, Organisation Owners manage this under Organisation settings → Capabilities.

How to use it

  1. Enable the capability in settings.
  2. Ask Claude to produce a file. Be concrete about format: “create an .xlsx with a monthly burn model” works better than “make a spreadsheet”.
  3. Claude runs code in the sandbox and returns a download link in the chat.
  4. Open it locally. Iterate in the chat and Claude rebuilds.

Upload + transform

Upload an existing file and ask Claude to transform it — extract PDF tables into a spreadsheet, convert a deck to a doc, pull CSV data into a chart.

Good to know

  • Supported outputs: .xlsx, .pptx, .docx, .pdf, plus Python scripts, PNG visualisations, and CSV/TSV output.
  • 30MB file size cap. Per file, for both uploads and downloads.
  • Network access is separate. By default on Team/Enterprise, network egress is off — the sandbox only has pre-installed packages. Owners can allow package managers, whitelisted domains, or full internet.
  • Project files are reachable. Files in a Project are available inside the sandbox.
  • Sandbox isolation + prompt-injection classifier. The sandbox is isolated; Anthropic also runs a classifier to flag attempts to trick Claude into exfiltrating data via the network. Disabling network egress removes that risk entirely.

What’s changed recently

  • 2026-04-18 — Page created from official help-center article.

Related

Sources