For questions that standard reports cannot answer, the assistant offers data profiling, a read-only query tool, and a secure analysis sandbox.
Business data analysis
The assistant can profile and analyse any record type the user can read. This is useful for questions like "How clean is our customer data?" or "What is the trend in new leads this quarter?"
| Analysis type | What it produces |
|---|---|
| Profile | An overview of each field: how many are empty, how many are unique, and basic figures. |
| Statistics | Averages, medians, spread, and quartiles for numeric fields. |
| Trends | Growth patterns over time, by day and by month, using a chosen date field. |
| Quality | A health check: duplicates, gaps, and consistency. |
| Correlations | Relationships between numeric fields, flagging strong ones. |
By default the analysis looks at up to 1,000 records and can be extended to 10,000 for larger studies.
Read-only database query
Administrator only. The database query tool is restricted to administrators because it bypasses normal record-level filtering. It is strictly read-only: it can only read data, never change it. Any attempt to modify data is blocked.
When an administrator needs a custom figure that no report provides, the assistant can run a read-only query. The tool refuses anything other than a read, blocks multiple statements, and caps the number of rows returned (default 100, up to 1,000).
The analysis sandbox
The assistant can run analysis code in a tightly controlled, isolated environment to fetch data and crunch it in a single step. This is the engine behind sophisticated answers like "Compare this quarter's basket size against last year, by region." The sandbox is restricted to a small set of trusted roles.
What the sandbox can and cannot do
Allowed
Reading data through safe, built-in helpers.
Standard data analysis with common analysis libraries.
Read-only database reads.
Blocked, by design
Any data change (create, update, delete).
File access on the server.
Network access to outside systems.
Interactive input and arbitrary code execution.
Charting inside the sandbox. Charts are made through the dashboard tools instead.
Safety limits
| Limit | Default | Range |
|---|---|---|
| Execution time | 30 seconds | 1–300 |
| Memory | 512 MB | 64–2048 |
| Processor time | 60 seconds | 1–300 |
| Output size | Capped at about 1 MB | Trimmed beyond this |
Administrators set these on the Security tab (section 8). Every code run is recorded in the audit log, including how long it took and whether it succeeded.