Execute code

When a question needs real computation, the agent can write and run code on your data, inside the conversation.

What you can do

• Run code in a sandbox (code_interpreter) — the agent writes and runs Python to compute, transform, or chart data.
• Bring files in, get files out — pass files as inputs and receive generated files back, such as a cleaned dataset or a chart.
• Automate small tasks — data crunching, format conversions, and one-off jobs that are easier in code than by hand.

Example ask

"Clean this CSV and chart monthly revenue."

What's coming

Sandbox mode — a dedicated workspace the agent controls to run code, perform analytics, and build interactive dashboards. See the Roadmap.