Use New Relic Autopilot

Once you've set up Autopilot, there are two ways to leverage it:

• On-demand chat: Ask ad hoc questions in the AI panel any time you want to investigate something. Autopilot is the default agent in the panel, so you no longer need to select it from a list before you start asking questions.
• Automatic analysis on alerts: Once you connect Autopilot to a workflow, its analysis is already waiting for you the moment you open a fired alert or issue—no need to ask anything.

This guide provides practical tips and best practices for getting the most out of Autopilot.

After you've set up Autopilot, you can start using it to analyze your environment and investigate incidents. This guide provides practical tips and best practices for getting the most out of Autopilot.

Usage tips

Follow these best practices to get optimal results from Autopilot:

Time range considerations

Autopilot is designed to prioritize live analysis and typically defaults to searching for events, issues, and entities within the last 24 hours. If you want to analyze something older, specify a targeted time window for analysis.

Query optimization

The agent uses tools designed to convert natural language into NRQL queries, which you can then review and run in the query builder. Autopilot is designed to analyze telemetry data and is not currently intended or designed to perform source code review or source code analysis. When referencing long conversations, limit ambiguous pronoun usage to help improve the accuracy of the results.

Always review suggestions carefully

Agents may not always act as you intend. For example, an agent may make a suggestion or initiate a process that it deems reasonable, but that may not align with what you had in mind. Due to this, we have incorporated a human-in-the-loop (you) for certain decision making in the Autopilot design and provide configurable controls for Autopilot access and permissions for more automated tasks. It is important for you to review suggestions carefully and validate those suggestions and next steps for your own context and circumstances.

Response time

Response times may vary depending on the scope and complexity of the investigation.

Give feedback

Use the thumbs-up and thumbs-down controls on any response, and add a comment. The most useful feedback states what you asked, what you expected, and what you received.

Review Autopilot's reasoning

While Autopilot is investigating, and after it responds, expand Show reasoning to audit what it did. This shows each step of the investigation, where information may have been pulled from, and whether the investigation step is still in progress. Completed steps will show a green checkmark. Autopilot flags steps that may have run into a problem, so you can spot where something went wrong.

New Relic AI Knowledge with Autopilot

You have the option to connect New Relic AI Knowledge to Autopilot so that its answers take into context your team's own documents, such as runbooks or other process documents, in addition to Autopilot's general knowledge. Once connected, Autopilot will search your connected content during an investigation.

With New Relic AI Knowledge enabled, you can:

• Connect Confluence retros, postmortems, runbooks, and custom files so the agent's answers can strive to reflect your team's actual processes and practices.
• Search past incidents that may closely match what you're seeing now, the steps that fixed it, and the team that handled it—so you spend less time investigating from scratch.
• Ask questions in plain language, such as "What are the standard triage steps for this alert?" or "Show me the runbook for a database connection limit exceeded error," and have content and context drawn directly from your own documentation.
• Type / in the Autopilot side panel to attach a specific knowledge document—a runbook, retro, or postmortem—to your prompt so the answer reflects that document.

For setup instructions and supported connectors, refer to New Relic AI Knowledge.

Use Autopilot in Slack

If an admin has connected your workspace to Slack, you can investigate without leaving the conversation. Mention the New Relic app (for example, @New Relic) in any channel where it's invited, and Autopilot replies in the same thread. You can:

• Ask why an alert triggered, check the health of a service, or analyze a recent deployment.
• Reply in an alert notification's thread to investigate it in context, without naming the entity.
• Ask follow-up questions in the same thread to dig deeper.

For workspace setup, identity linking, and the full usage guide, refer to Autopilot Slack integration.

Autopilot Memories

If enabled, Autopilot Memory extracts operational knowledge from your interactions—past failure patterns, effective workarounds, and infrastructure quirks—and uses it as a persistent knowledge base Autopilot draws on in future investigations.

For details on how memory capture works as well as associated access permissions and data retention, see Autopilot Memories.

Next steps