Live Reports in Azure SRE Agent now in public preview

Microsoft has introduced Live Reports in Azure SRE Agent, now in public preview. The feature lets operations teams describe a dashboard in natural language and have the agent generate a page with a fixed layout whose underlying data refreshes on every open. Microsoft positions it as a replacement for re-asking the same monitoring questions across tabs, scripts, and chat threads.

The core design choice is separating authoring from rendering. The agent builds the page once, complete with charts, thresholds, and column order, then freezes that structure. What changes on each load are the embedded tool calls that fetch fresh data. With no model in the render path, the page loads like a standard web page rather than a streaming chat response. Tokens are spent only when a report is authored or edited, not when it is opened. Reports opting into AI-powered interpretation are the exception, consuming tokens on each load and strictly opt-in per report.

When a team wants changes, it can ask the agent to edit the page in place, with prior versions preserved. Data sources are the connectors the agent already has access to, including MCP servers, logs and metrics, source control, work items, and internal services. Rendered output uses charting and grid libraries pulled from a CDN, producing interactive and sortable tables rather than static markdown.

The feature also supports two optional runtime behaviors. First, a report can send a focused prompt to a lightweight model at view time to classify alerts, group issues by root cause, or summarize a discussion thread. Second, a report can include action buttons that invoke write tools from connected connectors, such as acknowledging an incident or opening a work item. These buttons are limited to the exact tool names saved with that version of the report and cannot call arbitrary URLs or APIs.

That action capability is paired with a four-layer security model. Reports render in an iframe with sandbox="allow-scripts" but without allow-same-origin, so the page runs at an opaque origin with no access to session tokens or cookies. Network access is restricted, with all data requests routed through the Live Reports runtime, which validates each call against an approved tool manifest for that specific report version. Tool calls are constrained by per-version allowlists, and nothing executes until a human explicitly approves the tool descriptions presented at save time. Exports are sanitized to a static, offline-safe file.

The main benefit is consistency with freshness: a dashboard that looks identical every morning but shows current numbers, without the cost or variance of a model regenerating the entire view. Microsoft is asking users what operational views they end up rebuilding by hand. Live Reports is available via the Azure SRE Agent here.