Improvements
- Coding Agents
- GitHub Copilot and Codex usage is now ingested and broken down alongside the agents already supported
- The Overview and Cost tabs were redesigned: total cost now leads as a large hero number, and each metric carries a trend badge showing how it moved versus the previous period
- Logs: The log explorer offers horizontal scrolling for long lines when line-wrapping is off, instead of wrapping or truncating them. Scrolling a column keeps every cell in it in sync, so it’s easy to compare similar strings across rows
- Alerting
- Non-aggregate log alerts are now supported directly in the new alert editor
- “Edit in group settings” in the Notifications section links straight to the alert group’s Channels tab
- Discover Hosts: Host dashboards now show the EC2 name, instance ID, and other instance metadata alongside node metrics
- MCP: OAuth access tokens now last four weeks, so connected agents no longer need to re-authenticate every day
Fixes
- Dashboards
- A single panel whose instant query returned a values tuple instead of a single value could throw and take down the entire dashboard
- Public and shared dashboards didn’t load data, and Transpose didn’t work on shared dashboards
- Table panels built on aggregations without a group-by showed confusing labels — a stray
{}and a redundant metric column - Sub-second values with a seconds unit displayed as
0.00 seven when the chart showed non-zero spikes
- Alerting
- Some alert rules couldn’t be edited in the new UI
- The alert visualisation dialog overflowed its wrapper on narrower screens
- The Monitor empty state implied no alert rules existed when there were simply none in a firing state
- MCP: The hosted remote MCP server intermittently returned “session not found” on tool calls when running more than one replica
- General: The filters bar stretched too wide on large screens across several views, squeezing the content area and forcing horizontal scrolling