Webless

Execute one to four constrained, tenant-scoped Analytics v2 semantic queries for model reasoning, then run analytics-only Python over their trusted results in an isolated, network-restricted managed sandbox. Use this for calculations, comparisons, pivots, statistics, or display shaping that one semantic query cannot express. Python receives analytics_sources; every source exposes top-level columns, rows, truncated, rowCount, warnings, and provenance fields including applied filters and source plans. Declare every output in presentation.datasets and assign result to an object with exactly those dataset ids as keys; bind every visualization to one datasetId. A trusted source boundary error means the request cannot be retried unchanged: replace the crossing source query with one source per supplied range and rewrite Python for the new analytics_sources positions, while staying within the four-query limit. If safe reformulation is not possible, use separate analytics_query calls and report each range without claiming the computation completed; do not sum non-additive metrics. Histograms receive pre-binned category/count rows; heatmaps receive x and series dimensions plus one metric; funnels receive ordered stage/value rows. Raw SQL and free-form predicates are never accepted. This data tool does not render UI. When the user should see an interactive computed chart or table, call render_analytics with plan.operation=compute instead of calling this tool first.

analytics_compute

Run one tenant-scoped Analytics v2 semantic query for model reasoning and return typed rows, applied-filter provenance, and surface-neutral visualization specs. Compose only fields and filter operators advertised by analytics_catalog, use one semanticGroup per query, and obey each metric's timeRangeAlignment; analytics_compute can combine separate groups while preserving each source plan. If a trusted source boundary error supplies ordered time ranges, retry every supplied range as a separate analytics_query call without changing other query fields. Raw SQL and free-form predicates are not accepted. This data tool does not render UI. When the user should see an interactive chart or table, call render_analytics with plan.operation=query instead of calling this tool first.

analytics_query

Read the versioned metrics, dimensions, dimension filter operators, limits, semantic groups, timeRangeAlignment requirements, tenant-specific queryPlanning boundaries, and visualization capabilities supported by Analytics v2. Use this before analytics_query, analytics_compute, or render_analytics when choosing fields, filters, or time ranges. Obey queryPlanning.queryRule exactly: never submit a query that crosses an advertised source boundary; split the request into the specified non-overlapping ranges before calling another analytics tool. Metrics with different semanticGroup values must be submitted as separate semantic queries; analytics_compute and render_analytics with plan.operation=compute can combine their trusted results. Use exact UTC hour boundaries for utc_hour alignment and exact UTC day boundaries for utc_day alignment.

analytics_catalog

Execute and render one final Analytics v2 query or constrained computation in an interactive chart and table view. Use plan.operation=query for one semantic query and plan.operation=compute for calculations or display shaping. Call this exactly once for the final user-visible analytics view; do not call analytics_query or analytics_compute first when their only purpose would be to supply this renderer.

render_analytics

Restore a previously rendered analytics view from an opaque, authenticated capsule. This read-only tool is reserved for the attached Webless app; models must call render_analytics for new analytics requests.

replay_analytics