Ask Your AI to Analyse the Numbers: Sandboxed Code in Chat
Stop exporting CSVs to public AI tools. Upload, ask, and get a chart back. Sandboxed Python, no install, full audit trail.
The CSV-in-ChatGPT problem
Walk through any office and you will find people doing data analysis in public AI tools. Customer lists, sales exports, salary data, sometimes worse. They do it because the tools work and Excel is painful. They do not do it because they want to leak data.
Sandboxed code execution inside a governed assistant gives them the same capability without the data trip. Upload a file, ask a question, get a chart, all inside the perimeter you control. The audit log catches every step.
The real win is not the technology. It is that the people who were going to do data analysis anyway now have a sanctioned way to do it.
Three things people actually use it for
Question your spreadsheet
Upload a CSV or Excel file, ask a question in plain language, get the answer with the working shown. No formulas, no pivot tables, no "let me get back to you on that". Useful for the people who own the numbers but do not write code.
Quick chart from raw data
Ask for a chart, get a chart. Bar, line, scatter, distribution, whatever fits. The assistant picks a sensible visualisation and shows the code it ran, so a colleague who does write code can verify or adapt it.
Multi-file analysis
Compare two reports. Reconcile a sales export against a finance close. Stitch quarterly figures into a single view. The kind of one-off analysis that usually means an evening of Excel becomes a five-minute conversation.
What a typical exchange looks like
User: Here is the Q1 sales export. Which region underperformed compared to last year, and by how much?
Assistant: Reading the file. 12,400 rows, six regions. Comparing against last year's figures in your reference document.
Assistant: Northern region is down 18% year over year. Southern is flat. The other four are up between 4% and 11%. Here is the chart. I have included the calculation so you can verify.
The file never left the environment. The code that ran is visible. The result, the chart, and the audit entry all sit in the same conversation.
Why this is safe enough for Finance and HR to use
The capability would be irresponsible without the boundaries around it. Three properties make it a fair tool to put in front of departments that handle sensitive data.
Sandboxed execution
The code runs in an isolated environment with no internet access. It cannot reach your network, your databases, or any other system. When the conversation ends, the sandbox is gone.
Your data, your perimeter
The file you upload stays inside your environment. Nothing is sent to a third-party data analysis service. The assistant has access. Nobody else does.
Full audit trail
Every uploaded file, every line of code executed, every chart generated is logged with the user identity. If somebody asks "where did this number come from", the answer is in the dashboard.
Related articles
Ready to see it in action?
Schedule a personalised demo and see how the Plainsight AI Assistant fits your organisation.