AI data sharing

Modified on Fri, 17 Oct at 4:46 PM

A short guide to what data in Omniscope is shared with AIs. Written regarding Omniscope 2025.2.

You might be concerned about your sensitive company data being shared with another company, or even a foreign government - or used to train an AI model. We understand and share these concerns, and can assure you that Omniscope will never share data with an AI unless explicitly configured to do so.

Key points

1. AI integration is off by default

Omniscope will not speak to any AIs, unless you have explicitly:

Added and configured an AI provider (e.g. OpenAI, Azure OpenAI, a local model, etc.) with a valid API key.
Further configured an AI integration (such as Report Ninja or the AI Block) to use that provider and a specific model.

(See How to enable AI in Omniscope for setup instructions).

2. Omniscope only talks to the AI when expected

Even when AI providers are configured, Omniscope only communicates with them as part of an intentional or visible action. It never sends data silently in the background without your knowledge:

Integration	When AI communication happens
Report Ninja	Only when the Report Ninja right sidebar is showing. When the sidebar is collapsed, communication stops. The sidebar is collapsed by default for all reports except "Instant Dashboard" projects/reports.
AI Block	Only when you add an AI Block to the workflow, open it and start chatting. If not opened, no communication occurs.
Custom SQL	Only when you add a Database block to the workflow, select "Custom SQL", enter a question in "Auto-generate SQL by asking a question", and hit "Convert question to SQL".
Data Q&A View	Only when the Data Q&A has been added to a report tab, and that tab is open. This includes using the Data Q&A report block preset.
Workflow Ninja	Only when the workflow ninja sidebar is showing. It is collapsed by default.
AI Completions block	Only when you add an AI Completions block to the workflow, connect it to input data, and execute the workflow - at which point data is sent for completion processing.

3. Omniscope only shares data according to configured data sharing levels.

Each AI provider now has a configurable Data Sharing level, which determines how much (if any) dataset information can be shared with that provider.

This allows an administrator to set appropriate limits for different providers, for example stricter limits for cloud-based third parties, and more permissive limits for privately hosted open-weight models.

To configure, in the Admin app, select "AI Settings", select a provider and navigate to "Data sharing level".

Data sharing levels (from most to least permissive):

Level	Description
Full datasets	Unrestricted sharing of entire datasets, to allow full processing by the AI model. All fields, records and cell values can be shared. Allows complete AI processing.
Sample records	Limited sharing of up to 10 sample records with all cell values in those records. Helps to explain the data, without the ability to retrieve specific records.
Sample cells	Sharing of cell values for a given field, without being able to relate those values to specific records or cell values in other fields. Allows precise understanding of individual fields, such as range and formatting of specific names to support accurate filtering.
Schema only	Sharing of field-level metadata only, not including cell values. Includes field names, data types, description/source/group annotations and unique value counts. Supports the understanding of the data model without revealing actual cell values.
Table identifiers	Sharing of table-level metadata only, not including any identifying field, record or cell data. For example, the filename of a CSV source file, a database table name, record/field counts.
None	No information about the dataset can be shared. Only chat text entered by the user and information about the current Omniscope configuration excluding settings that refer to field names.

By default when a new configuration is added the data sharing is set to Full datasets.

4. Each integration has a minimum data sharing level

Each integration has minimum data sharing level that it can operate with. If the configured level is lower than required the integration will be unavailable for that provider. If it meets the minimum, but is below optimal the integration will gracefully degrade with reduced functionality.

Integration	Minimum level	Ideal level	Behaviour when level is lower
Report Ninja	Schema only	Sample records	Works at schema level, but without sample data the ability to create filters and create useful dashboards will be degraded.
AI Block	Schema only		Works normally. No degradation.
Custom SQL	Schema only		Works normally. No degradation.
Data Q&A View	Full datasets		Unavailable if lower. Requires access to all cell values to answer questions and create charts/diagrams.
Workflow Ninja	Table identifiers	Sample records	Works normally. If schema or sample records are allowed it can include schema or preview data. If lower, some features are unavailable.
AI Completions Block	Full datasets		Unavailable if lower. Requires full dataset access for completions.

5. You can lock Omniscope down

An administrator can define and enforce a central AI policy by configuring providers and integrations at the root folder level and disabling subfolder overrides:

Configure the AI providers and their data sharing levels in the root folder.
Untick "Allow subfolder override" to prevent users from using alternative providers.
Limit access to the root folder to prevent users from modifying your central policy.

6. You should carefully consider which AI providers to use, and what data sharing is appropriate

Here are some examples of how an you might configure data sharing with a set of 3 providers:

OpenA/Azure OpenAI - "Schema only". Provides metadata without sharing any meaningful data.
Custom privately hosted LLM such as llama.cpp -"Full datasets". Trusted environment with unrestricted use.