Data handling
What AI features send, what is retained, and how to restrict both.
This page states plainly what leaves your team when AI features are enabled, and how to narrow it.
What is sent
| Feature | Sent | Not sent |
|---|---|---|
| Mapping suggestions | Field names, types, up to 50 sampled values per field | Full tables |
| Dataset summaries | Field names, types, dataset and source names | Values |
| Ask your data | Your question, field names and types, the generated query | Result rows |
| Anomaly notes | Aggregate statistics — row counts, null rates | Individual rows |
Sampled values are real customer data
A 50-row sample of a contacts table contains 50 real email addresses. If that is unacceptable for a dataset, turn sampling off for it — the control is per dataset.
Retention
| Data | Retained | Where |
|---|---|---|
| Suggestions and your decisions | Life of the team | Chartizer |
| Ask history — question and query | 7 / 90 days by plan | Chartizer |
| Result rows from Ask | Not retained | — |
| Samples sent to the model provider | Not retained by the provider | — |
Chartizer does not train models on your data, and its providers are contractually barred from training on data submitted through the API.
Restricting it
Turn off value sampling for a dataset
Open the dataset → AI → uncheck Allow value sampling. Names and types are still used; suggestion quality drops but nothing from the rows leaves.
Exclude a dataset entirely
Exclude from AI features removes it from suggestions, summaries, and Ask. It becomes invisible to the assistant.
Turn a feature off team-wide
Settings → AI disables any feature for everyone. Members cannot re-enable it.
Mark fields as sensitive
Fields tagged sensitive are never sampled, never named in an Ask query, and are masked in
previews — even when the dataset otherwise allows sampling.
{
"ai": {
"sampling": false,
"excluded": false,
"sensitive_fields": ["email", "phone", "national_id"]
}
}
Tag sensitive fields as soon as the dataset appears
It is far easier than auditing later, and the tag follows the field through mappings into every project that reads it.
Regions
| Plan | Processing region |
|---|---|
| Free, Pro | EU (Frankfurt) |
| Enterprise | EU or US, fixed per team |
Enterprise teams can pin processing to a region and receive a signed subprocessor list.
Local agents
An agent keeps credentials and query execution on your infrastructure. It does not make AI processing local: if sampling is enabled, the agent still returns sampled values to Chartizer, which sends them to the model provider.
An agent is not an AI boundary
Teams often assume an agent means nothing sensitive leaves. It does not. For a database that must stay fully private, run the agent and disable sampling on its datasets.
Compliance
- Data processing agreement available on Pro and Enterprise.
- Subprocessor list published, with 30 days’ notice of change.
- Audit log records every AI action with actor and timestamp — see Access requests.
- SOC 2 Type II report available under NDA for Enterprise.