Operational Health

- AI Signals: Token counts in AI Explorer, My Work, and Livestream tables now display in compact form such as 7.11B, with the full value on hover, making large numbers easier to scan and compare.

• AI Signals: A new AI VCS Repository Branch dimension groups AI spend by the git branch the work was done on. Use it alongside the repository dimension to attribute AI cost to individual features, tickets, and pull requests.
• AI Signals: Cost attribution is more accurate across providers. GitHub Copilot credits are now allocated by model and user login, model author is correctly identified for Databricks-hosted GPT, Azure speech, and Cursor Composer models, and LiteLLM gateway events now carry the provider deployment and billing organization so proxy-routed spend can be traced to the account that served it.
• AI Data Plane: You can now skip AI classification for specific requests by sending the x-cloudzero-classification-opt-out: true header. Use it on high-volume agent workloads that already declare their context, to avoid paying for a classifier call on every request.
• Monitors: Monitors built on a dimension that was later deleted in Dimension Studio now show a warning in the list and on the Overview tab, so you can fix monitors that can no longer match cost.
• Monitors: Kubernetes Namespace and Workload filters in Monitors and Views now offer an Idle option, so you can scope a monitor or view to idle Kubernetes cost. Monitors scoped this way show "Idle" in their details.
• Monitors: A View in Explorer link in monitor details opens Explorer in a new tab with the monitor's group by, filters, and cost type applied, so you can investigate the underlying spend.
• Monitors: The open incident count on each monitor now links to the Incidents tab filtered to that monitor, and incidents can be filtered by monitor and status.
• Monitors: An incident's Activity tab now records each time its spend moves between Ongoing and Recovered, so you can follow the full lifecycle of an incident.
• Explorer: Copying a currency cell in the Results table now pastes the raw number, such as 525127.90, so you can drop values straight into Excel without cleanup.
• Views: Excluding a dimension that other dimensions are built from now asks for confirmation and lists the affected dimensions, and Explorer explains when a click has no further drill-down available.
• Budgets: Edit budget now asks for confirmation before a save removes an existing Budget or Forecast, so you do not lose its alert history by accident.
• Cloud Connections: AI telemetry connections are now excluded from health grading unless you set an expected ingest rate, so they no longer show false unhealthy warnings.
• Azure Connection: Enterprise Agreement connections are more self-service. Billing periods now include tax so CloudZero totals match your Microsoft invoice, billing currency is detected automatically with a mismatch warning on Test Connection, and each ingest run reports health checks for billing scope, cost reports, and subscription tags with guidance on how to resolve gaps.
• GCP Connection: New billing and recommender connections now default to a 24-hour expected ingest rate, which better matches Google's data delivery and reduces false delay warnings. Existing connections are unchanged.
• Windows Collector: The collector now stores its keys in Windows protected storage, so credentials stay encrypted and readable only by the signed-in user.
• Windows Collector: A system tray icon now shows live delivery status, so you can confirm at a glance that AI usage data is reaching CloudZero.

- AI Signals: The Token view in AI Explorer now loads its chart reliably instead of intermittently showing "Unable to load chart data."

• AI Signals: The usage livestream no longer falls behind during weekday peak hours. Since September 30 it could lag by up to 45 minutes; usage now appears in near real time.
• AI Signals: Amazon Bedrock model names with regional prefixes, ARN suffixes, or vendor prefixes, including models from xAI, Z.ai, and MiniMax, now resolve to their base model name. Pricing applies correctly to these models and Bedrock billing lines up with your AI telemetry.
• AI Signals: Inference events delivered more than once are now counted a single time in hourly token and cost totals, so spend and token counts for affected gateways match your provider. The fix applies to newly processed data; contact support if you need earlier periods reprocessed.
• AI Signals: Negative filters now return correct results across Overview, AI Explorer, and My Work.
• AI Signals: An unmapped AI provider no longer creates a duplicate connection when the organization already has an active connector for that vendor, so cost is not counted twice.
• Monitors: Syncing a ticket that was deleted in ServiceNow now completes cleanly instead of failing repeatedly, so deleted incidents no longer trigger a stream of repeated error alerts.
• Monitors: The AI User Monitor now reliably detects its required AI dimensions even if a dimension display name is changed, so incidents are no longer skipped in affected organizations.
• Monitors: Incident analysis is more reliable. It now completes for large organizations during peak periods and is preserved when the tag and ownership lookup times out, so you still receive cost findings. AI incident summaries name a user or team as the driver only when it accounts for most of the increase.
• Explorer: Month to Date and Year to Date ranges no longer show an end date before the start date on the first day of a period, so the header matches the date picker and the chart loads.
• Views: Limited access users whose role uses an exclude rule now see their permitted views, including the Default view, instead of an empty list.
• Optimize: Nightly insight analysis now completes for every insight, so open and addressed savings totals are calculated and published each night.
• Optimize: Database Savings Plan recommendations now reflect the most recent AWS recommendation for each payer account. After you purchase a plan or your usage changes, the suggested commitment and estimated savings update instead of staying at an earlier peak.
• Optimize: Older Generation Instances savings are now calculated from actual pricing of comparable newer Intel and AMD instance types, replacing a flat 15% estimate. Each recommendation shows both options with matching vCPU and memory, so you can choose an upgrade that lowers cost rather than raises it.
• Optimize: Claude Code context growth recommendations now match your Anthropic billing resources and appear in the Optimize list. Previously, findings were generated but could not be linked to the underlying users, so none were displayed.
• Analytics: Alerts now keep running in two cases that previously stopped them: when the owner would have been deactivated by the inactive-user cleanup, and when the owner belongs to multiple organizations and switches between them.
• Analytics: Restricted folder visibility now also applies when a folder or its dashboards are opened by direct link, so only users in permitted groups can view, edit, copy, or add content to it.
• Cloud Connections: Row counts and data volumes on connection ingest statistics are updating again after they stopped on September 1. Only these statistics were affected; cost data was ingested normally throughout.
• Cloud Connections: OpenAI cost data now includes projects owned by other OpenAI organizations on the same enterprise account, so reported spend reconciles to your invoice.
• Cloud Connections: New Relic costs for data ingestion are now complete for August 25 through September 10, when New Relic sent split ingestion records that could cause billable usage to be omitted.
• AWS Connection: Connection health now shows the failure when an AWS data ingest run fails, instead of keeping the last status, so you can spot and act on problems sooner.
• Azure Connection: Invoice reconciliation for Microsoft Customer Agreement accounts is now correct early in the month, so temporary refunds no longer appear against spend while invoices are still being finalized.
• Databricks Connection: The Dynamic Account Pricing setting now displays correctly and can be turned on or off from the connection detail panel on every connection, including older ones, so you can adjust pricing behavior without contacting support.
• Databricks Connection: Account pricing failures caused by a fixable configuration issue, including a service principal missing a required permission scope, now return a clear message that names what to correct instead of a generic error.
• Datadog Connection: Container and infrastructure costs are no longer counted twice when Datadog's hourly attribution reports both a product name and its alias, so affected totals now match actual spend.
• Oracle Cloud Connection: Report downloads now retry automatically when Oracle drops the connection, so a brief network interruption no longer fails an ingest run.
• Snowflake Connection: Re-ingesting April through June 2026 no longer double-counts Cortex Code costs. Those months were correct as originally ingested; only a re-ingest was affected.
• Snowflake Connection: Connections no longer show a degraded status on the first day of the month when the new month has no data yet, so status reflects real ingestion problems only.
• macOS Collector: GitHub Copilot usage from bring-your-own-key models is now retained when GitHub reports a zero charge, so this usage still appears in your AI telemetry.