Release Notes

With an A/B test you take one benchmark, answer it on two sides, and compare them question by question. A side is a semantic model and the AI that answers for it, so you can compare a semantic model against a copy of it you are optimizing, one semantic model answered through two different AIs, or two data agents that differ only in their instructions. Side A supplies the questions, and BI Pixie computes the correct answers once per run from side A's semantic model before either side is asked, so both sides are scored the same way. Before the run, BI Pixie reports how many of those answers side B's semantic model agrees with, without asking any AI. The result leads with the hypothesis you stated: one BI Pixie suggested comes back confirmed, rejected or inconclusive, and one in your own words is left for you to judge. Every question is paired side by side. The Free and Standard plans include one A/B test run that does not draw on your benchmark runs. On the Pro and Enterprise plans, each side that runs uses one benchmark run, and side B's semantic model becomes a tracked item. Learn more.

BI Pixie now assesses your Fabric data agents as well as the semantic models they are grounded on. Each data agent receives its own AI Readiness score from 0 to 100, with every finding explained in Microsoft's own published words, and BI Pixie caps the score by the readiness of the semantic models the data agent answers from. BI Pixie proposes the instructions, grounding, published description, and runtime change that resolve the findings, and writes nothing until you approve each one. When you choose, BI Pixie asks your AI about the checks it cannot answer on its own. The first assessment of each data agent is free on every plan. Learn more.

BI Pixie now prepares your Power BI semantic models for Copilot in Power BI and for Fabric data agents. Assess a semantic model across 12 AI Readiness dimensions with the findings ranked by the fields your people actually query, optimize it with drafts you approve before anything is written, and create benchmarks that test the AI against correct answers computed from your own data. The first assessment of each semantic model is free, for up to 500 semantic models on every plan. Learn more.

BI Pixie Workload for Microsoft Fabric is now generally available in the Fabric Workload Hub. Any Fabric tenant can add BI Pixie from the Workload Hub with the standard permissions — the preview tenant setting is no longer required — and the full experience is now available: dashboard, engagement metrics, embedded surveys, RLS auditing, AI assistants, maintenance notebooks, OneLake data residency, and Bring Your Own Storage. Learn more.

The BI Pixie MCP Server is now generally available. Connect AI assistants like Claude, GitHub Copilot, Microsoft Fabric data agents, and Azure AI Foundry to your BI Pixie data and ask how your Power BI reports are really used, in plain language. Strictly read-only, with numbers that match your BI Pixie Dashboard. Learn more.

BI Pixie is now live in the Microsoft Fabric Workload Hub as a public preview. Any Fabric tenant can add BI Pixie directly from the Workload Hub and start tracking how Power BI reports are used without leaving Fabric. Learn more.

Connect AI assistants like Claude, GitHub Copilot, Microsoft Fabric data agents, and Azure AI Foundry to your BI Pixie data and ask how your Power BI reports are really used, in plain language. Strictly read-only, with numbers that match your BI Pixie Dashboard. Learn more.

BI Pixie began running natively inside Microsoft Fabric as a workload, with a BI Pixie item living right in your Fabric workspace. Launched in private preview and since superseded by the public preview. Learn more.

BI Pixie is now available as a cloud service. Sign up, connect your Power BI workspaces, and start tracking user engagement in minutes, with no setup or infrastructure to manage. Learn more.

Enterprise customers can deploy the full BI Pixie data pipeline in their own Fabric workspace, backed by a Fabric Lakehouse with Direct Lake connectivity, incremental refresh that loads only new telemetry, and zero-ID CI/CD deployment. Learn more.

A new way to streamline BI Pixie with your existing CI/CD processes. Automatically export items from selected workspaces and update them back with tracking pixies. Works great with Test stages of Power BI deployment pipelines. Learn more.

BI Pixie now knows what types of visuals exist and whether they are built-in, certified or unknown. This knowledge is incorporated into the BI Pixie Dashboard (Azure version only).

Previous version of the instrumentation didn't recognize the live connection reports if the external model was not a Power BI semantic model.

Detect data leakage due to flaws in RLS implementation. Learn more.

See new dashboard page: "Endorsement and Data Sources".

Track and analyze export to Excel/CSV events in Power BI. Learn more.

You can combine Power BI Activity Logs with BI Pixie's unique instrumentation technology to bring a new perspective into your BI producers and consumers.

Collect Power BI Activity Logs to bring a new perspective into your BI producers and consumers. Includes Paginated report views and more activities.

Automatically add customized surveys in your instrumented reports to collect user feedback.

With the new integration, you gain insights into new consumers' activities.

The automation was aborted when TMDL semantic models had calculation groups.

When the heatmap was activated, there was inaccurate count of clicks.

Enhanced auditing capabilities allows you to analyze what data users see and what data they select. Learn more.

New Data Auditing page to analyze the data users select in visuals.

You can now collect performance data from Azure Log Analytics. Learn more.

Reports with live connections and no external DAX measures were not instrumented.

When you use measures instead of columns as the fields in Drill-through pages, BI pixie automation was aborted.

When data visuals were grouped the X and Y coordinates in the heatmaps of the dashboard were not accurate (They were relative to the group coordinates).