Privacy & Context Modes
Three modes controlling what schema metadata reaches an LLM, with a per-module breakdown of exactly what each one sends. No mode sends raw data rows.
Privacy & Context Modes give you control over exactly what schema metadata is sent to an LLM when Edilitics uses AI features. Three modes are available across Integrate, Transform, AskEdi, and Visualize. The modes behave differently per module because each module uses AI for a different purpose. No mode in any module sends raw data rows to an LLM.
The Three Modes
Private sends the least context of the three. In AskEdi, Transform, and Visualize it protects column identity: real column names are never sent, and columns are replaced with anonymized labels such as col_1, col_2. In Integrate it works differently, because generating a description for a column requires knowing what that column is called: real column names are always sent, and Private instead withholds your Focus Sector and frequent values. Structural metadata (table name, data types, DQ statistics) is included in every mode so the LLM can produce useful output.
Balanced sends real table names and real column names alongside DQ statistics. No frequent values or sample data. This is the default for everyday analysis where column names are not sensitive.
Full Context sends everything Balanced sends, plus frequent values observed in each column. In AskEdi, it also includes the previous LLM reply as context for follow-up questions. Use this when you need the deepest reasoning and column names and value patterns are not sensitive.
Integrate
AI features in Integrate use context to generate column metadata, descriptions, and insights for your schema.
| What is sent | Private | Balanced | Full Context |
|---|---|---|---|
| Table name | Yes | Yes | Yes |
| DQ statistics | Yes | Yes | Yes |
| Focus sector | No | Yes | Yes |
| Frequent values | No | No | Yes |
Column names and data types are always included in Integrate because they are required for metadata generation. Focus sector adds domain context to improve the relevance of generated descriptions. Frequent values help identify categorical patterns in your columns.
AskEdi
AI features in AskEdi use context to generate SQL queries and interpret results. Column identity is the primary privacy concern here because column names appear directly in generated queries.
| What is sent | Private | Balanced | Full Context |
|---|---|---|---|
| Table name | Yes | Yes | Yes |
| Anonymized column names | Yes | No | No |
| Real column names | No | Yes | Yes |
| Data types | Yes | Yes | Yes |
| Column insights | Yes | Yes | Yes |
| User role | Yes | Yes | Yes |
| DQ statistics | Yes | Yes | Yes |
| Frequent values | No | No | Yes |
| Previous LLM reply | No | No | Yes |
In Private mode, column names are replaced with anonymized labels such as
col_1, col_2. Generated SQL queries use these labels. You can still run
and interpret results, but the queries will not use your real column names.
Visualize
AI features in Visualize use context to suggest chart types, map columns to axes, and generate dashboard layouts. The payload structure is identical to AskEdi, without the previous reply toggle.
| What is sent | Private | Balanced | Full Context |
|---|---|---|---|
| Table name | Yes | Yes | Yes |
| Anonymized column names | Yes | No | No |
| Real column names | No | Yes | Yes |
| Data types | Yes | Yes | Yes |
| Column insights | Yes | Yes | Yes |
| User role | Yes | Yes | Yes |
| DQ statistics | Yes | Yes | Yes |
| Frequent values | No | No | Yes |
Transform
AI Transform Suggestions use context to propose the next operations in a pipeline. Transform uses the same three modes as Visualize, and the mode is chosen each time you generate a set of suggestions.
| What is sent | Private | Balanced | Full Context |
|---|---|---|---|
| Table name | Yes | Yes | Yes |
| Anonymized column names | Yes | No | No |
| Real column names | No | Yes | Yes |
| Data types | Yes | Yes | Yes |
| Column insights | Yes | Yes | Yes |
| User role | Yes | Yes | Yes |
| DQ statistics | Yes | Yes | Yes |
| Frequent values | No | No | Yes |
The operations already applied in your pipeline are sent with the request in every mode, so suggestions build on what you have rather than repeating it.
Frequently Asked Questions
Related Docs
Data Encryption
How all stored data is encrypted at rest using per-workspace key derivation.
Audit Logs
Every action logged with user, timestamp, and description. Privacy mode changes are tracked.
Bring Your Own Key (BYOK)
Use your own API keys for OpenAI, Sarvam, Anthropic, or Google DeepMind and control provider-side logging.
Need help? Email support@edilitics.com with your workspace, job ID, and context. We reply within one business day.
Last updated on
Data Encryption
How Edilitics encrypts every piece of data it stores using per-workspace key derivation. What is stored, what is never stored, and how encryption works.
Audit Logs
Every workspace action is logged with the acting user, timestamp, and a plain-English description. What is tracked, who can access logs, and how to export.