Autopilot 2.6

What's new

Compute limits

Compute limits give Organization Managers a level of budget control over Autopilot. Now, New Relic administrators can set daily and monthly budgets in Advanced Compute Capacity Units (Advanced CCUs) to help govern spend on investigations and automated alert triage. A separate sub-budget for automated investigations helps keep alert-driven activity and manual work each within the budget you set, so you can plan Autopilot costs with confidence.

• Set daily and monthly budgets: Organization Managers define budgets in Advanced Compute Capacity Units (Advanced CCUs).
• Reserve a budget for automated investigations: Set a separate daily and monthly sub-budget for automated work, such as investigations started by alert webhooks and workflow notifications.
• See your usage in the chat panel: A status pill in the chat context bar shows where your organization stands.
• Warnings before you reach the limit: From 75% of your budget, Autopilot adds a warning to its responses that your organization is approaching its daily or monthly limit.
• Message when the limit is reached: At 100%, new requests are paused and Autopilot explains the next steps: contact your Organization Manager to raise the budget, or try again after the daily reset at 00:00 UTC.

For setup steps and official details, see Autopilot compute limits.

Increased successful investigations and compute limits to manage your budget

New features

• Set compute budgets for Autopilot (Preview): You can now set daily and monthly compute budgets to help you stay in control of costs during busy periods such as alert storms.
• See Autopilot work through each stage: The thinking display now shows distinct planning and summarization steps, with a status message that updates as Autopilot moves from one stage to the next.
• Improved NRQL guidance: Autopilot is better at building valid NRQL queries, with improved handling of faceting, grouping and query structure.

Improvements & fixes

• Root cause is less likely to default to a recent deployment: Investigations of error rate, latency, and synthetic monitor alerts better handle deployment history rather than preferencing a recent deployment.
• Recalled information is kept separate from current evidence: Autopilot can now better parse causes of prior issues from the current investigation evidence.
• Empty results are flagged: Fixed an issue where a step that returned no results or low confidence could be treated as complete. Autopilot is more likely to call out when a part of an investigation comes back empty.
• Fewer stalls on repeated lookups: Fixed an issue where an investigation could stall when Autopilot needed to look up the same data a second time.
• Enhanced protections for data fencing: Autopilot includes enhanced safeguards for fencing and handling data only versus agent instructions during investigations.
• Fast investigations error gracefully: During fast investigations Autopilot now reserves time to summarize what it found, so you are more likely to get a useful answer.