Extended Views

Views now save everything: time range, granularity, chart type, series count, and dimension visibility, not just filters and group-bys. They live inside the expression builder, so you can select, create, and manage views right where you build your queries. A side panel lets you browse all views with an inline preview of the rendered table and chart before saving. Set any view as your default, share it with a clean, readable URL that survives renames, and scope it to specific dimensions so CloudZero can optimize how it computes cost data behind the scenes.

API Key Scopes That Say What They Grant

Every API key scope now describes what it grants, and scopes are grouped by the job a key does, from least to most access, so you can give an integration only the access it needs. Keys that hold a retired scope are flagged with its replacement, and existing keys keep their scopes.

Monitors and Incidents in the API and MCP

The public API and the CloudZero MCP server now let you list, create, and update monitors and incidents, manage notifications, and change an incident's status, so you can automate alerting workflows. New API key scopes grant monitor and incident access on their own, so automations don't need broader permissions.

Anthropic Enterprise Skill Costs

Anthropic Enterprise cost is now split by skill for each user and day, so you can see which skills drive spend and allocate it accurately.

Ticket Field Mapping

Admins can now map CloudZero dimensions and tags to Jira or ServiceNow fields, per project and ticket type, so tickets created from CloudZero arrive prefilled with the cost context your teams need. You can also edit or remove the details CloudZero fills in.

macOS Collector

A new desktop app runs capture in the background with a tray icon, start at login, and a status popover that shows delivery status at a glance. Administrators can manage it through device management, including disabling AI provider sign-in and deploying corporate root certificates, and changes apply without relaunching the app.

Cursor and Claude Code Coverage

Cursor activity from the native app, SDK, and Tab completions is now captured, so you can see AI usage across every way your team uses Cursor. Claude Code activity is now classified by task, as Copilot usage is, so you can see what your teams use AI coding for.

Data Changes

Azure Connection: Microsoft Customer Agreement invoices now reconcile to the credits shown on the invoice instead of commitment drawdown, so spend for commitment-funded accounts no longer shows phantom refunds or reads lower than billed. Non-USD invoices now reconcile as well, applied to closed billing periods as they are processed.
Datadog Connection: Cost data for accounts with very large volumes now finishes ingesting instead of timing out, so previously stale data resumes updating. Attribution tags now appear as a single tags field going forward.
AI Signals: Usage from Codex CLI is now ingested as one event per agent turn, sub-agents included, so Codex model and token usage appears in your AI cost data instead of being dropped.
AI Signals: AI usage on conversations that work on a Salesforce account is now attributed to that account, so you can group AI cost by customer account. The account is recorded only when its ID appears in the conversation, and attribution applies to new activity going forward.
Snowflake Connection: Credits for five Cortex AI services, including AI Inference and AI Guardrails, are now priced at their own published rates, so AI spend is no longer overstated by about 16%.
Elastic Connection: Cost data now stops at the last completed hour, so projected spend for future hours no longer inflates monthly totals or triggers budget alerts early.
OpenAI Connection: Fast mode usage is now classified as Priority Tier, and GPT-6 cache writes are recognized, so model names and tiers in GenAI cost views are accurate.

Improved

- Optimize: GCP Cloud Storage buckets with high cross-location read charges are now flagged, with a savings estimate for each.

AWS Connection: Every billing period in a CUR 2.0 export is now ingested, so history that AWS backfills for newer connections now appears automatically without a support request.
GCP Connection: Organizations can now separate GCP resource labels from Resource Manager tags into distinct dimensions, so a tag with the same key no longer hides a label's cost data.
GCP Connection: Data export failures now show the specific cause, such as a timeout or existing files, and monthly export retries are safe, so you can resolve ingestion issues faster.
Access Controls: The default group can no longer be deleted, and group role changes are validated before saving, so a failed update cannot leave a group half-changed or without a valid role.
CloudZero MCP: Assistants can now list your connected ticket projects, look up a Jira project's required fields, and fill them in when creating incident tickets. Missing fields are named in the error, and retrying a ticket request no longer creates duplicates, so tickets are created correctly the first time.
AI Signals: The Anomalies Overview card now has an Active and Resolved toggle, and opens on Resolved when none are active, so you no longer see an empty card.
Monitors: The redesigned monitor form and Monitors tab are now available to everyone, with Active Alerts and Created By columns. Filters and Group By are optional side-by-side buttons, and the threshold preview chart shades a normal range around expected spend and labels each anomalous day, so you can see exactly where an alert would trigger.
Monitors: The Alerts list now loads in pages with a Load more button, with search, filters, and sorting running on the full alert set. Date range presets such as Last 90 days and Last 12 months cover longer history, alert cards show a "% above baseline" badge, and your Cards or List choice is remembered.
Monitors: Saving a monitor that duplicates an existing configuration now shows a View existing monitor link, and the API returns a conflict response with the existing monitor's ID, so automation that retries after a lost response can reuse it. Duplicate detection also compares cost impact thresholds at full stored precision.
Monitors: AI incident analysis now runs on a newer Claude model, finishes faster, and holds up during large bursts of incidents. Summaries state only what the cost data supports, can name the Jira epic behind a cost increase, and refresh when anomalous days change or billing data is restated. Re-sending an incident's current status no longer adds activity log entries or Jira updates.
AI Signals: Hovering over an abbreviated token count on the Overview now shows the full, comma-separated number, so you can see exact usage without leaving the page.
Explorer: The date range now stays fully adjustable while a View is applied, so you can change the View's default dates freely, just like its filters and group-bys.
Views: You can now set weekly and monthly trend notifications when creating or updating views through the public API, so managing many views as code no longer resets notification settings.
AI Signals: Released data plane images are now signed and ship with software bills of materials, so you can verify image provenance before deploying.
Cursor: Spend cards now flag incomplete usage instead of showing zero, and the app tracks which counters Cursor actually reported, so you can tell unknown values from genuine zero usage. Local spend reads from captured usage, with Agent and Tab activity grouped under Cursor.
Telemetry Streams: Queries on optimized telemetry now stay available while the data is being rebuilt and after the rebuild completes.
AI Signals: A single deployment guide now covers both intake modes, with a setup guide for each supported source. The documentation also states that classification experiments export numeric embeddings derived from prompts and replies, never text, so security reviews have an accurate picture of what leaves your cluster.
Analytics: Saved dashboard filters now support drill paths up to five levels deep, so you can follow spend from department through team, epic, employee, and activity.
Explorer: An Expand All button opens every top-level row when two or more group-bys are active, and Expand 2 Levels opens two levels from any row. Each level shows the top 50 items with the rest rolled into an Other row with an exact count, and a new % of Parent column shows each item's share of the row above it. Available in Labs.
Explorer: Admins can now choose which dimensions Global View includes or hides, so Explorer shows only the dimensions relevant to your organization while Global View is active.
Dimension Studio: Dimensions in CostFormation now accept an optional description of up to 500 characters, so your team can see what each dimension represents. Description-only edits do not reprocess your cost data.
AI Signals: Filters, reports, and the AI overview chart now use AI Model Author and AI Cloud Platform in place of the retired AI Model Provider, and saved reports that used the old dimension load as Model Author.
AI Signals: Activity from the CloudZero AI collector is now recognized and classified when the data plane runs in OpenTelemetry mode, so each model round trip is attributed to the user and conversation that produced it.

Fixed

- Optimize: Insights and the cost summary for potential savings and anomalies now load reliably, including for users with limited access during heavy load.

Azure Connection: New connections open correctly right after setup, connections created before late 2022 can be edited again, and temporary Azure Advisor outages no longer cause recommendation ingestion to fail with a false error alert.
Anthropic Enterprise Connection: Connections no longer fail for most of the day when skills data for the previous day is not yet published, so your other cost data keeps syncing.
Ticket Integrations: Ticket syncing now retries quietly when a Jira site is temporarily restricted during an Atlassian migration, and a ServiceNow permission denial returns a clear error instead of a server failure. Incident tickets created through CloudZero MCP now choose the right integration for organizations with both Jira and ServiceNow.
Monitors: The Overview card now follows the selected date range instead of always showing the last 30 days, so it matches the rest of the page.
Explorer: Favorites and History filters on empty or unset values now display as "(none)" instead of an internal placeholder, so saved views are easier to read and search.
Views: Invalid characters in a view name are now flagged as you type, with the allowed characters listed, so you can correct the name before saving instead of hitting a generic error.
Monitors: The Spend Status column in the Alerts list is now wide enough to show "Spend Elevated" in full, so you can read each alert's status at a glance.
Explorer: Views with a relative date range, such as last 30 days, now include the most recent complete day, so the dates match the equivalent date range preset.
Explorer: Your role's default cost type now always appears in the cost type selector, so you can switch back to it after exploring another cost type.
Monitors: Incident analysis now recovers when the AI requests data with an empty filter value, and incidents whose analysis fails or times out still appear in your list with a clear title built from the monitor name and filters.
Monitors: Forecasts now use each monitor's configured cost type, such as amortized cost, so alerts and exclusions for those monitors reflect the same cost view you selected.
Optimize: The GCP multi-region storage recommendation now reflects each bucket's current billing, so recreated regional buckets are no longer flagged and one-time early deletion fees no longer inflate savings.
AI Signals and macOS Collector: Certificates from a compromised certificate authority are no longer trusted by the data plane image or the collector's exported certificate bundles, and the AI Signals egress collector includes fixes for several critical and high severity networking vulnerabilities.
SSO Integrations: Group Sync can no longer be turned on for Google OpenID Connect integrations, a setting that caused Google to reject every login for the organization. The message now points to Google Workspace SAML for syncing groups.
Dimension Studio: Publishing a custom dimension that references a deleted telemetry stream is now rejected immediately, so a stale stream reference no longer blocks your cost data from updating.
AWS Connection: Daily ingestion now completes for organizations with more than 1,000 AWS accounts, so Trusted Advisor data refreshes across every account instead of stopping partway.
Explorer: Choosing Global View from the View selector now returns Explorer to Global View and shows "Global View" in the selector, even when a Group By is already applied.