Admin Settings → Code controls whether assistants and workflows can run code — the engine behind data analysis, chart generation, file conversion and computation in chat.
Enable and choose the engine
| Option | Effect |
|---|---|
| Enable Code Execution | Master switch for running code. |
| Enable Code Interpreter | The assistant-facing code capability in chats. |
| Code Execution Engine | Which engine runs generated code (built-in runner or a connected execution service such as Jupyter). |
| Code Interpreter Engine | Engine selection for the chat interpreter specifically. |
| Jupyter URL / Auth (token or password) | Connection details when a Jupyter-based backend is used. |
| Code Execution Timeout | Maximum seconds per code run — prevents runaway jobs consuming the cluster. |
| Code Interpreter Prompt Template | Optional custom instructions guiding how the interpreter behaves; empty uses the default. |
Code execution is privileged:
Generated code executes inside the platform's sandbox/execution environment. Keep execution on managed infrastructure, restrict it by permissions, and combine it with Guardrails. If your threat model forbids arbitrary code, leave it disabled — chat remains fully functional for everything else.
Practical guidance
- Timeouts: start conservative (tens of seconds); raise for data-analysis groups with justification.
- Who needs it: data analysts, engineers, document-pipeline builders — not necessarily everyone.
- Monitoring: code runs appear in usage statistics like any other activity — Usage.
Troubleshooting
| Message | Meaning |
|---|---|
| "Failed to load code execution settings" | Configuration service unreachable — check cluster health. |
| Interpreter errors for users | Engine URL/auth wrong or the execution service is down. |
| Users can't see the capability | Enabled here ≠ permitted for their group; check Permissions. |