New Relic has launched Ground Truth CLI, a command-line tool that pulls its observability investigations into the terminal for developers and AI agents. The pitch is “headless observability”: reading production telemetry without opening a dashboard. The tool is a limited preview. Its category is already busy.
What New Relic announced
New Relic Ground Truth CLI, augmented with an Autopilot API, lets engineers and AI agents query services, logs, alerts and changes from the terminal. It connects to Autopilot, NerdGraph and memory APIs. Teams can define recovery criteria, automate checks, and capture investigation evidence for a handoff. New Relic announced it at New Relic Now, its virtual event.
The CLI is a limited preview, not a release
The headline says New Relic “introduces” the CLI. The availability note says something narrower. It is “planned to enter limited preview” on 8 October 2026. Ground Truth and Autopilot are generally available; the CLI is not. Teams weighing adoption should read the preview status before planning.
The received version also carries a re-dated dateline. The wire announcement ran on 6 October from San Francisco. This version is dated 8 October from Bangalore. That is a syndication variant, not a second launch.
Headless observability is already a crowded field
New Relic frames headless observability as the “next generation.” Competitors shipped first. Datadog offers Pup CLI, an AI-agent-ready command line with more than 200 commands across 33-plus products and its own agent mode. Grafana ships gcx, an agent-first CLI for Grafana Cloud, Enterprise and open source. Honeycomb has hccli. Alibaba Cloud has obz-cli.
The direction is real, and the market agrees on it. New Relic is joining that direction, not setting it.
Autopilot recommends, it does not act
The release leans on AI agents driving investigations. New Relic’s own documentation sets limits on that. Autopilot “recommends actions but does not take them.” It does not create alerts. It does not write back to external systems such as ServiceNow or Confluence. A person reviews and acts.
That is a sensible design for production. It also qualifies the autonomous framing. An agent that reads telemetry and proposes a fix is useful. An agent that changes production on its own is a different proposition, and New Relic has not shipped that.
The security question the release skips
A CLI that reads production telemetry is a privileged client. It needs access to Autopilot, NerdGraph and memory APIs. Independent analysis notes the limited preview does not detail the connection model, and recommends treating the tool as privileged and scoping it to least privilege.
New Relic’s own MCP documentation stresses OAuth, role-based access control and administrative toggles. Granting AI agents read access to live production data also splits responsibility: agents surface findings, engineers validate changes. The release does not discuss any of this.
What to watch
Whether the CLI reaches general availability, and how its access model is governed. Whether New Relic’s MCP route and its CLI route converge into one surface. And whether “headless observability” becomes a real category or stays a marketing frame for terminal access.
Editor’s Note
This analysis draws on the New Relic Ground Truth CLI release and its wire version, New Relic’s documentation for Ground Truth, the MCP server and Autopilot, Datadog’s Pup CLI, Grafana’s gcx, Honeycomb’s hccli, independent coverage from DevOps.com and Help Net Security, and reporting on New Relic’s October 2026 announcements and leadership change. Where a figure is a company’s own or an estimate, we say so.