Dashboard snapshots
Freeze a dashboard exactly as it looked at a moment in time — the underlying data is captured and saved with the snapshot, so it keeps rendering even after the live data ages out of retention or the incident is long over. Each snapshot is pinned to its captured time range, gets a shareable link, and shows up in a snapshots list where you can copy the link or delete it.
This is built for incident work: capture the state during an outage and drop the link straight into an RCA report or postmortem, so everyone sees the same charts you did without needing dashboard access or the right time window. Snapshots can also be created and managed through the MCP server, so the AI Assistant can snapshot what it found mid-investigation and hand you a link.
A Kubernetes tab on your services
Discover → Services gets a Kubernetes tab, with an Events sub-tab, so you can see the Kubernetes workloads behind a service — and the events they emit — without leaving the service page.
Improvements
- AI Assistant
- Slack replies now stream in as the assistant works, instead of arriving all at once
- Exception details link straight into the AI Assistant to continue investigating
- Alerting: Alert health history is now part of the new alerting experience. The 14-day health view — a grid of hourly status dots — used to live on a separate screen and was scoped only to an alert group. Now every firing alert in Alert Monitor has a History tab, and you can view it at rule scope for finer granularity as well as at group scope.
- General: Chart legend interactions are now consistent across every surface — single-click a series to isolate it and view it on its own, and Cmd/Ctrl+click to drop that series and keep the rest
Fixes
- Applications (RUM)
- Android warm and hot app-start types were never emitted, because the startup latch fired only once per process
- App Not Responding (ANR) errors were invisible in the Errors tab because “Errors by Type” grouped on the wrong attribute
- The Screen Load Time panel was always empty because it queried the wrong span name
- An Android security exception could be thrown during OkHttp network-type detection
- Dashboards
- Bar charts crashed on bar click and on drag-to-zoom
- Bar chart panel previews showed a double tooltip
- Empty timestamps at range boundaries were skipped
- Discover Services
- Double-clicking a dependency in the service map opened the wrong page
- Service Catalog gRPC client latency incorrectly summed across peers
- Alerting
- The log preview in Slack alert notifications failed to load in some cases
- Alerting was delayed by per-alert log-line fetching
- AI Assistant: Markdown and tables rendered incorrectly in assistant responses in both the browser and Slack, and overly long prompts now map to a clear context-limit message