The Sandbox Model
SNcode never runs AI-generated code behind your back. Every script follows the same review-first flow — you see a plain-language description and the exact code before anything touches your instance.
Review-first execution
The AI proposes; you approve. When SNcode generates a script, it presents three things you control, in order:
- 1
Describe
A natural-language summary of what the script will do and which tables or records it touches. - 2
Inspect
The full generated code, exactly as it will run — nothing hidden, nothing minified. - 3
Execute
You run it in an isolated sandbox with full control. Nothing executes until you say so.
Why it matters
- Large language models can produce incorrect or incomplete code — a human review gate is mandatory, not optional
- You keep an auditable record of what was proposed versus what was run
- Admins can connect a client instance knowing no code executes without explicit approval
AI output is not automatically trusted
All AI-generated code, scripts, and recommendations must be reviewed by a qualified human before execution or deployment. The sandbox model exists to make that review the default, unavoidable step.
Execution history
Script executions and their outputs are recorded locally on your device, so you can trace exactly what ran against which instance and when — useful for audits and for resuming work across sessions.
Next
The sandbox reasons over more than tables. Next, see how attaching CSV, PDF, and Excel files lets the AI transform your own data — still behind the same review gate.