CloudZero Academy
Changelog
CloudZero Status
Back to All

AI Signals

July 2nd, 2026

AI Signals Explorer now supports double group-by, so you can partition AI spend by two dimensions at once, such as Model by User or Service by Region.

The AI Signals livestream Activity column is now visible by default, so you can see what's driving your real-time AI spend at a glance.

A "Last 14 days" quick-range preset is now available in the AI Signals date picker, making week-over-week comparisons one click away.

Improved

  • AI Signals: Organizations sending AI telemetry through OpenTelemetry can now include custom business dimensions, so AI spend can be grouped and allocated by team, product, cost center, or any category you define.

  • AI Signals: Service and machine traffic from API keys is now identified by key alias instead of showing as "Unknown," so every caller in your AI usage summaries is attributable.

  • Telemetry Streams: The API and CSV upload form now accept zero values for allocation streams and negative values for metric streams, supporting use cases like refunds and zero-activity records without errors or workarounds.

  • AI Signals: Selected filter values now float to the top of the value picker, making it easier to review and adjust active selections.

  • AI Signals: The Explorer filter builder now supports free-text contains/does-not-contain and null operators, matching the filtering capabilities available in the main Explorer.

  • AI Signals: Estimated cost totals in AI Signals Explorer now include a disclaimer noting that costs are based on on-demand pricing, matching the pattern used in the main Explorer.

  • Optimize: The recommendation details flyout now labels savings as "Potential Monthly * Savings" so the time period is immediately clear.

  • Kubernetes: A Region column is now available on the cluster list page, so you can distinguish clusters that share a name across different regions.

Fixed

  • AI Signals: Cache token costs for LiteLLM-based integrations are now calculated at the correct cached rate, resolving an issue where cached input tokens were priced at the full rate and overstating costs.

  • AI Signals: AI telemetry billing records are no longer generated for cloud-provider-billed AI services, preventing double-counted costs for usage already captured through AWS, Azure, or GCP connections.

  • AI Signals: Custom dimension names no longer collide with built-in AI dimension identifiers, so customer-defined dimensions resolve correctly.

  • AI Signals: The cost-over-time chart now renders hours with zero spend as zero-cost bars instead of collapsing the time axis, and date picker selections align to clean UTC day boundaries regardless of local timezone.

  • AI Signals: Removing a filter no longer causes it to reappear when drilling into a table row.

  • Datadog Connection: OAuth authorization and reauthorization flows now complete successfully, resolving a redirect issue that prevented connecting or reconnecting Datadog integrations.

  • Explorer: Kubernetes efficiency metrics now remain visible when filtering by cloud provider, region, or account.

  • Views: Updating a saved view with a renamed dimension no longer fails with a permission error.

  • Telemetry Streams: Allocation calculations no longer include records with missing target values, preventing incomplete records from skewing cost distribution.