AI Signals Overview

- Default Landing Page: You can now choose which page CloudZero opens to from Personal Settings, set a default Explorer view, and pick your preferred date and number formats, so the platform adapts to the way you work. CloudZero also remembers your Explorer right rail layout between sessions.

• Explorer: Period-over-period comparison now uses a single start-date picker with auto-derived range and shows focused columns (Total Cost, Previous Period, Cost Difference), reducing setup from six parameters to three.
• Explorer: Right-clicking a cell value in any table now shows "Filter" instead of "Add to search," matching the action it actually performs.
• Cloud Connections: Connection status now reflects what CloudZero actually knows for each billing period, showing a brand-new connection as pending rather than errored and flagging a phase as blocked only when ingest is truly stuck.
• GitHub Connection: Per-line-item GitHub tag values such as user login and email are now kept as line item properties, so you can allocate and filter Copilot spend by individual user even when several share a resource.
• Snowflake Connection: Snowflake connections can now reference another Snowflake connection as their rate source, so blended-rate cost data flows into the correct billing context.
• Navigation: Hovering the tenant switcher in the collapsed side navigation now shows the name of the tenant you are signed into, instead of a generic label.
• Budgets: You can now click a month in a budget's History chart to load that month's daily spend, giving you day-level detail for closed months instead of a single total.
• Optimize: A new recommendation flags Claude Code sessions that repeatedly replay large amounts of context after being paused and resumed, showing the cache-read cost this drives and where to reduce it.
• Optimize: You can now select all matching recommendations at once for a bulk status update, instead of checking rows one at a time.
• macOS Collector: Opening the app while it is already running now brings up its settings window, giving you clear confirmation it is running even when its menu bar icon is hidden.
• AI Signals: The Overview shows cost and token trends in side-by-side charts with hourly, daily, or weekly granularity, an anomalies card that opens incident details in place, and click-to-filter across every tile. Every panel offers CSV export, a percentage column with totals, shift-click to isolate a legend series, and a 60-day date range picker. You can drill into unassigned costs and open any row directly in AI Explorer.
• AI Signals: AI Explorer now supports free-text filter operators, a Cost/Tokens toggle with token-native columns, and a unified Caller field that consolidates user, API key, and service account identities. Anthropic ephemeral cache write tokens now appear in the usage breakdown. Cost figures are labeled "Estimated Cost" with a tooltip explaining they reflect list pricing.
• AI Signals: The Control Plane device flyout now shows the device's display name, parent collector, hostname, first-seen date, and ownership details. Custom tags sent with AI traffic automatically become spend dimensions, a context enrichment card explains what enabling enrichment unlocks, and usage from Google Gemini and Vertex AI is tracked as its own service.
• Monitors: A step-by-step monitor builder lets you create and edit alert conditions from a drawer, with a live 30-day cost preview that shows what your condition will detect before you save. Available in Labs.
• Monitors: Incident, anomaly, and recommendation views now lead with an AI summary of the likely cause. The feed spans 60 days with a 7-day hourly drill-down, the Spend column is sortable, and each incident links to its originating monitor with description and Slack channel visible in the details panel. AI spend anomalies can now attribute a spike to a specific work item or repository.
• Monitors: You can now edit the threshold and notification settings on system monitors, so you can tune sensitivity without recreating the monitor.

- AI Signals: The Models count on the Overview tab now reflects active filters, matching the Top models panel, so a filtered view no longer shows the organization-wide model total.

• AI Signals: Agent-orchestration rollup spans are no longer billed alongside their child model calls, fixing a double-count that inflated token totals for agentic workloads.
• AI Signals: The AI Explorer date picker now anchors to UTC day boundaries and preserves preset precision, so a 7-day preset always covers exactly seven full days.
• Monitors: Filters that referenced a deleted dimension could not be cleared; deleted dimension filters are now surfaced and removable.
• Monitors: The start date in a Cost Sentinel incident email now matches the start marker drawn on the incident chart, so the email and chart agree on when an incident began.
• Monitors: Creating or updating a monitor no longer fails or times out when its cost-data validation runs long, so monitors save reliably even with complex custom-dimension filters.
• Monitors: The minimum cost impact setting now saves correctly, gates on the cost overage above the expected baseline rather than the day's total cost, and the anomaly detection threshold floors at twice the robust standard deviation, reducing false positives on low-variance cost series.
• Explorer: The comparison panel now stays open after loading and the Apply button enables correctly when the comparison period matches the base period length, so you can adjust without reopening the panel.
• Explorer: Row expand carets in a two-dimension group-by now work correctly regardless of how many parent rows are loaded.
• Views: Saved filters now carry into the new View form correctly, so stale filters from deleted dimensions no longer linger when you create a view.
• Views: 'Is null' and 'is not null' filters now carry over correctly when you create a view from Explorer, instead of being changed to a blank condition.
• Cloud Connections: OpenAI cost imports now handle very large daily responses and recover when the provider rejects certain breakdowns for older dates, so cost data keeps flowing instead of the entire import stalling.
• Snowflake Connection: Snowflake Cortex AI charges are no longer double-counted and are now priced against the correct rates, so your Snowflake AI spend is reported accurately.
• GitHub Connection: Per-seat license costs are now attributed to each individual user instead of collapsing onto one account, so GitHub seat spend is allocated correctly across all license types.
• Optimize: AWS Trusted Advisor recommendations no longer fail to load for your whole organization when a single flagged resource returns fewer details than expected.
• Optimize: Recommendations that break down cost by pricing term now generate correctly, instead of failing to load for that grouping.
• Optimize: Realized savings now load correctly when filtered by a specific recommendation or insight, instead of returning an error.
• Optimize: Claude Code cost recommendations now cover sessions from the standard interactive login and attribute them to the right person, so far more of your organization's usage qualifies for savings guidance.
• Optimize: Insight Chart Details no longer fails to load cost data when the insight was created from a two-dimension group-by.
• Budgets: The pace line in the budget flyout now reflects the actual elapsed time in the period instead of always reading 100%, so you can once again compare spend against how much time has passed.
• CloudZero MCP: Container usage requests now run on a dedicated per-organization warehouse, so queries return in seconds instead of queuing behind other work and timing out.
• Telemetry Streams: You can now map a telemetry stream to CloudZero-defined customer dimensions such as AI User, which previously returned an error when you tried to save.
• macOS Collector: The usage section of the menu bar popover no longer appears empty when you are signed out or while data is still loading, so you always see meaningful content.
• macOS Collector: Usage from Claude Opus 5 and Sonnet 5 is now priced instead of showing no cost, so spend from these models is captured accurately in your AI usage.
• Settings: The organization message toast now clears when you switch organizations or when an admin disables it, instead of persisting from the previous org.