Privacy Modes
How Private, Balanced, and Full Context modes control what AskEdi sends to the AI, and how each mode affects response accuracy.
Privacy modes control exactly what context AskEdi sends to the AI when it generates a query and synthesises a response. You select a mode when creating a chat, and it applies to every question in that conversation. The choice is a trade-off: more context enables more accurate responses, less context keeps more of your schema private.
No raw data rows are ever sent to the AI in any mode.
The privacy mode is locked when a chat is created. It cannot be changed mid-conversation. To use a different mode, start a new chat.
The Three Modes
What the AI receives:
- Real table name
- Anonymized column identifiers:
col_1,col_2,col_3(real names are hidden) - Column descriptions
- Data types
- DQ statistics: null counts, distinct value counts, value ranges
What is not sent:
- Real column names
- Most frequent values
- Previous analysis results
How queries work:
AI generates the query with anonymized names
The AI writes a query using col_1, col_2, etc. It never receives real column names.
AskEdi rewrites the query
Before execution, AskEdi substitutes the anonymized identifiers back to real column names.
Query executes against your source
The rewritten query with real column names runs against your database or file.
Analysis view shows both queries
The Analysis view displays the anonymized query the AI produced and the rewritten query side by side, so you can verify the exact column mapping applied.
What to expect from responses:
Private mode gives the AI structural information about your columns but not their names. It relies on column descriptions to understand what each anonymized field represents. If your column descriptions are complete and accurate, Private mode can produce precise results. If descriptions are missing or vague, the AI has less to work with and responses may be less specific about which fields drove a finding.
Private mode accuracy depends on column description quality. Before using Private mode, verify that your columns have meaningful descriptions in AI Column Insights.
When to use:
Use Private mode when your column names themselves carry sensitive meaning, for example fields like national_id, credit_score, or churn_risk_flag. Private mode lets the AI reason about structure and semantics without knowing what the columns are called.
What the AI receives:
- Real table name
- Real column names
- Column descriptions
- Data types
- DQ statistics: null counts, distinct value counts, value ranges
What is not sent:
- Most frequent values
- Previous analysis results
- Raw data rows
What to expect from responses:
Balanced mode gives the AI your real column names and structural statistics. It understands which columns are sparse, what ranges numeric fields cover, and how many distinct values categorical columns have. This is sufficient for most analysis questions and produces accurate, well-grounded responses without exposing the actual content of your data.
When to use:
Balanced is the default mode and the right choice for everyday internal analysis where column names are not sensitive. The real column names and DQ statistics give the AI enough context for accurate query generation without sharing what values your data actually contains.
What the AI receives:
- Real table name
- Real column names
- Column descriptions
- Data types
- DQ statistics: null counts, distinct value counts, value ranges
- Most frequent values for each column
- The previous analysis result (to support follow-up questions)
What is not sent:
- Raw data rows
What to expect from responses:
Full Context mode gives the AI the richest picture of your data. Knowing the actual top values in each column lets it reason about the real content of categorical fields, for example which product categories exist, which status values are in use, or which regions appear most often. This matters most for questions where the answer depends on understanding what values are present, not just what columns exist.
The previous analysis result being included as context means follow-up questions build naturally on prior findings without you needing to restate them.
When to use:
Use Full Context when categorical distributions in your data affect the accuracy of the analysis, or when you are working through a multi-step investigation where each question builds on the last. It is not recommended when data minimisation is a governance requirement, since frequent values from your columns are sent to the AI provider.
Mode Comparison
| Context sent to AI | Private | Balanced | Full Context |
|---|---|---|---|
| Actual table name | Yes | Yes | Yes |
| Real column names | No | Yes | Yes |
| Anonymized column names | Yes | No | No |
| Column descriptions | Yes | Yes | Yes |
| Data types | Yes | Yes | Yes |
| DQ statistics | Yes | Yes | Yes |
| Most frequent values | No | No | Yes |
| Previous analysis result | No | No | Yes |
| User role and focus sector | Yes | Yes | Yes |
| Raw data rows | Never | Never | Never |
Frequently Asked Questions
Related Docs
Analysis View
See the anonymized and rewritten queries side by side in Private mode.
AskEdi
How a chat works, response types, and how to start a conversation.
Decision Intelligence
How AskEdi handles decision-type questions and what the response includes.
Need help? Email support@edilitics.com with your workspace, job ID, and context. We reply within one business day.
Last updated on