Finance Apps
A new Apps section in the sidenav gives finance teams purpose-built tools for AI spend. CapEx Classification breaks AI spend down by department, activity, repository, and work item, with drill-down and CSV export for capitalization reporting. Spend Projections shows year-to-date totals and monthly projections by model and author, so you can compare the current pace with last month. Apps require AI Signals with AI data flowing, and access follows AI Signals permissions. Reach out to your CloudZero account manager if interested
Session Duration Settings
Administrators can now set how long users stay signed in to CloudZero, with one setting for the web app idle timeout and one for the maximum session length of AI tools connected through the MCP server. Set both under Settings > Organization Settings > Session Duration to match your security policy or keep AI tools connected longer. Organizations that never change these settings keep the defaults. See the session duration documentation for details.
Data Changes
- Telemetry Streams: The Telemetry Streams API now creates a stream on the first replace-period write, the same way sum and replace already do, so you can load data without creating streams ahead of time.
- Azure Connection: Reservation line items now include the reservation name shown in the Azure portal and a unique label, so you can map reservations to custom dimensions and reconcile with Azure. These properties are added to reservation rows as cost data is processed.
Improved
- Cloud Connections: Connection operational health now includes each connection's latest reported ingest status, so you can tell user-actionable failures from other errors and prioritize the connections that need your attention.
- Optimize: AI-generated summaries for Azure and GCP recommendations now include the resource's cost data, so the savings context in each summary reflects actual spend.
- Monitors: Usage-based alerts now show the hour they started, in UTC, on alert cards, the alerts table and incident details, so you can pinpoint when a usage spike began.
- SSO Integrations: The Azure setup forms now flag a Secret ID pasted as the client secret, or a secret pasted as the client ID, and say what to copy instead, so errors are caught before saving.
- Cloud Connections: The "Use fixed IP egress" tooltip now lists CloudZero's fixed egress IP addresses and links to the provider's setup guide, so you can update your allowlist without leaving the connection form.
- Cloud Connections: The delete confirmation now states that deleting a connection permanently removes its historical cost data, and suggests pausing instead to stop new ingestion while keeping history.
- Monitors: Incident details for AI telemetry monitors now show an hourly AI cost chart scoped to the affected dimension, so you can see when spend changed and which dimension drove it.
- Monitors: AI telemetry incidents now list the sessions that contributed to the spike, with each session's user, cost impact, share, and activity. Each share is calculated against spend in the anomalous hours only, so you can see exactly what drove an unexpected increase in AI spend.
- Monitors: AI telemetry spike detection now counts an hour as active only above $0.50 of spend, reducing false alerts on series with sporadic small charges. Series that never exceed $0.50 in an hour are no longer checked for spikes.
- Collector: Version 1.2.0 for macOS adds work context for Cursor, records Claude chats on claude.ai, labels each Claude record as Chat, Cowork, or Design, and retries usage that failed to send once the server is reachable again. Administrators can also turn on content capture, which is off by default, and point the Collector at their own data plane address, both through a device management profile.
- AI Data Plane: Captured conversations now ship with their matching usage records to your data plane, never to CloudZero, and are deleted from the device once delivered, keeping content under your control.
Fixed
- Views: Dimensions created after a view is saved now stay visible when the view hides specific dimensions, so new dimensions no longer disappear from Explorer without anyone choosing to hide them.
- Cloud Connections: Validating an AnyCost connection with rejected credentials now returns a clear, actionable error instead of a generic server error, so you can see which field to correct.
- Datadog Connection: Cost data collection now retries automatically after a dropped network connection, so a brief Datadog interruption no longer leaves a connection failing and cost data delayed for days.
- Datadog Connection: Ingestion no longer fails when a usage day falls outside Datadog's 15-month history limit; that day is skipped and the rest of the run completes, so the connection stops showing a recurring error.
- Optimize: The recommendations list now loads reliably for organizations with very large result sets, so pages no longer time out when many recommendations are returned at once.
- Azure Connection: Resource counts are now correct for connections that cover multiple subscriptions in one tenant, where resources of the same type and region were previously shown as terminated and re-created each day. The correction applies going forward.
- User Settings: The My Roles card is now shown to every user on Personal Settings, so you can see which roles you belong to even without permission to manage role settings.
- Explorer: A copied URL for an applied View now keeps its group by, cost type and filters when opened in a new tab, so shared links show exactly what you saw.
- General: Clicking anywhere in a table column header now sorts that column, not only the title text.
- Optimize: The savings field in filters, group-by, and chart pickers now reads "Potential Savings", matching the recommendations table, so the filter is easy to find.
- Monitors: Incident analyses now finish with a full summary when the AI model asks for one more investigation step, instead of ending early with a placeholder title.
- Monitors: Anomaly incidents now show Recovered only after two days of billing data have arrived following the last anomalous day, so they no longer flip back to Ongoing when delayed data lands.
- CloudZero MCP: Cost, dimension, telemetry, and monitor tools now work for organizations with very large dimension catalogs, which previously caused these tools to fail on every call.
- Collector: One rejected usage record no longer causes the other records in its batch to fail, and the app now waits before retrying during a server outage.
- Collector: AI usage from requests that finish after the Collector's first pass is no longer dropped, so reported usage now includes them. Applies going forward.
- Collector: Claude chat turns from Claude Desktop and claude.ai in the browser are now recorded with their model and conversation, filling a gap where some chat usage was missing. This usage is recorded but not priced. Applies going forward.
- Collector: ChatGPT usage now keeps loading when OpenAI changes fields the collector does not use, and diagnostics name the exact field if a value cannot be read.
- Optimize: RDS Reserved Instance recommendations at the payer account level now appear, so previously missing opportunities are included and the total estimated savings for this recommendation type is higher for affected organizations.
- Optimize: EC2 Reserved Instance recommendations now show only the term with the higher savings for each instance group, instead of counting 1-year and 3-year alternatives together. Your total estimated EC2 Reserved Instance savings will be lower as a result.

