Privacy
What the LLM sees, and how long Doram keeps your data.
You choose the LLM provider, and your data never trains a model. Requests are processed in the region your workspace was created in.
What the LLM sees
| When | It can see | It never sees |
|---|---|---|
| Building the second brain | Table and column names, column profiles, query shapes without values, masked sample values, document excerpts | Rows of your data, credentials |
| Answering a question | The question, business names, and a short preview of your own results | Credentials, hidden columns, other people's data |
Sensitive values, such as passwords, government IDs, card numbers and dates of birth, are removed before anything is sent. Email addresses are partly masked.
The full result of a query goes to you, not to the model.
How long Doram keeps things
| What | Kept for |
|---|---|
| Questions and answers | 12 months. Database rows aren't stored with them. |
| Cached results for dashboards | 24 hours |
| Schema snapshots | 12 months |
| Audit log | 2 years |
| Everything, after you close your workspace | Deleted within 30 days |
When a person's data has to be erased, it's removed from the evidence and samples in every version of the graph.