Code Execution

Admin › Settings › Code Execution

Configure the runtime for AI-generated code.

This page configures the code execution engine (enable + Pyodide/Jupyter), and that engine is shared by both user-manual execution and the agent’s Code Interpreter (auto-execution).

Code Execution settings tab — Configure the code execution engine (Pyodide/Jupyter) — shared by user-manual execution and the agent Code Interpreter

Code Execution vs Code Interpreter

	Code Execution	Code Interpreter
Behavior	User manually runs AI-generated code blocks	AI auto-generates and runs code
Trigger	Click the “Run” button on a Python code block	When the AI decides data analysis is needed
Activation	Code Execution enabled	Execution engine set to Jupyter + chatting inside a data analysis project

A feature that lets users directly run AI-generated code blocks in chat.

Setting	Description	Default
Code Execution enabled	Code execution feature ON/OFF	Enabled
Execution engine	Code execution environment	`pyodide`

Execution Engines

Pyodide
Jupyter

Run Python inside the browser. WebAssembly-based — doesn’t affect the server.

Pro	Con
No separate server needed	Some native packages unsupported
No security risk	Large-data processing limits
Ready to use immediately	No filesystem access

Run code on a Jupyter server. All packages in the server environment are available.

Setting	Description
Server URL	Jupyter server address
Auth method	None, Token, or Password
Auth token/password	Authentication info
Timeout	Execution time limit (seconds)

Jupyter execution runs arbitrary code on the server, posing serious security risks. Use only in trusted environments.

Code Interpreter

A feature where the agent auto-generates and runs code for data analysis, chart generation, etc.

This tab has no separate settings for it — the Code Interpreter uses the code execution engine (Pyodide/Jupyter) configured above as-is.
It only works inside a data analysis project. It can’t be enabled in a regular conversation.

For Code Interpreter to work, both conditions must be met:

The execution engine in this tab is set to Jupyter — Code Interpreter runs in a Jupyter kernel, so Pyodide won’t work
The user is chatting inside a data analysis project

Without a configured Jupyter server, creating a data analysis project is blocked outright.

Code Interpreter isn’t a toggle users flip. Inside a data analysis project, the AI invokes it on its own when it decides analysis is needed. There is no matching item in agent capability settings either.

​Code Execution vs Code Interpreter

​Code Execution

​Execution Engines

​Code Interpreter

Code Execution vs Code Interpreter

Code Execution

Execution Engines

Code Interpreter