Usage data

Fabric Desktop by KratosBI can send a small daily summary of which actions were used, how often they failed and how long they took. It is off until you turn it on in Settings > Usage data. Your organization can turn it off, or on, for everyone. This feature is in Preview.

What is sent#

One record per day, after the day is over. Settings > Usage data > See what's sent shows the exact text waiting to go. Every field is listed here:

Field	Example	What it is
`day`	`2026-10-17`	The day the counts are for
`install_id`	`8f2c1e7a-5b0d-...`	A random id made on this PC. It isn't derived from you, your account, the PC or your tenant. New install ID replaces it
`build`	`58a298b1c2d3`	The app build
`channel`	`stable`	Stable or Preview
`windows`	`10.0.26100`	The Windows version
`python`	`3.12.7`	The engine's Python version
`features`	`{"run_sql": {"count": 9, "errors": 0, "seconds": 4.2}}`	For each app action that changes or runs something: how many times it ran, how many failed and the total seconds. The names are the app's own action names, the same ones the audit trail uses
`errors`	`{"HTTPError": 1}`	How many failures of each kind. The kind is the error's type name only

What is never sent#

Your data, files, tables, previews or query results.
Code, SQL, notebook contents or anything you typed.
Names of workspaces, lakehouses, tables, files, notebooks or entities.
Error messages, paths, URLs or ids from Fabric.
Your name, email, account, tenant or organization.
Read-only actions such as browsing workspaces or previewing a table. They aren't counted at all.

Where it goes#

The records go to Azure Monitor Application Insights over HTTPS, as one custom event named daily_usage per day.

By default they go to KratosBI's Application Insights. A build without a KratosBI destination has the switch but sends nothing, and Settings says so.
An organization can send them to its own Application Insights instead. Set telemetry_connection_string in the policy file. Nothing then goes to KratosBI.

When it is sent#

The app sends finished days once each time it starts, in the background.

A day that fails to send waits for the next start.
Unsent days older than 30 days are deleted.
Turning usage data off deletes everything waiting to be sent.
The headless agent never sends usage data.

The counts are kept in %LOCALAPPDATA%\FabricDesktop\telemetry\usage.json until they are sent.

For IT admins#

Set these in %ProgramData%\FabricDesktop\policy.json (Enterprise install):

Key	Value	Effect
`telemetry`	`false`	Off for everyone. People can't turn it on
`telemetry`	`true`	On for everyone. People can't turn it off. Use this with your own `telemetry_connection_string`
`telemetry_connection_string`	`InstrumentationKey=...;IngestionEndpoint=https://...`	Sends usage data to your own Application Insights resource

A connection string that isn't valid sends nothing, and Settings > Help and about lists it as ignored.

Allow the IngestionEndpoint host of the connection string through your proxy (for example westeurope-5.in.applicationinsights.azure.com). Without it, nothing is sent and nothing else is affected.

For maintainers#

A build sends to KratosBI only when the repository variable TELEMETRY_CONNECTION_STRING holds an Application Insights connection string. CI writes it into the engine's engine.json. Leave the variable unset to ship builds that never send. The code is in sidecar/fabric_desktop/telemetry.py.