RadarTranslation site

Website Description: Radar is a Kubernetes visualization and AI-powered operations tool developed by Skyhook that consolidates resource topology, event timelines, Helm, GitOps, and MCP interfaces into a single dashboard, supporting both on-premises open-source deployments and cloud-based team collaboration.

What is Radar?

When troubleshooting Kubernetes issues, the challenge often isn’t a missing command, but rather that logs, resource relationships, and change history are scattered across different locations. Designed for developers, SREs, and platform teams, Radar brings these clues together in a single interface and enables an AI assistant to investigate specific issues. It is not a cloud server management dashboard, nor does it require teams to first purchase Skyhook’s deployment platform.Official Product Documentation

Radar is developed by Skyhook’s platform engineering team and operated by KoalaOps, Inc. (operating under the Skyhook brand). The official website lists CEO Nadav Erell, CTO Eyal Dulberg, and CPO Roy Libman as the company’s leaders. The project originated from the team’s own need for cluster visibility and subsequently evolved into the open-source Radar and the team collaboration product Radar Cloud; both can be selected independently of the Skyhook deployment platform.Team and Product Background

Core competencies

• Understanding Resource Relationships: Use the topology and resource browsers to understand workloads, services, and related objects, rather than switching back and forth between multiple command outputs.
• Chained Events and Configuration: Incorporate information such as the event timeline, Helm, and GitOps into the troubleshooting context to help engineers track changes that occurred around the time of the failure.
• Providing cluster context to AI: It includes a built-in MCP service that can connect to compatible clients; there’s no need to choose between your existing operations interface and AI workflows.
• Conduct an investigation into the issue: Local mode can invoke existing Claude Code, Codex, Cursor, or OpenCode CLIs; the Cloud offers managed investigations. Investigation results include evidence and coverage; when data is insufficient, the system may provide an “undetermined” conclusion rather than directly interpreting “nothing found” as a sign of good health.Functional Documentation,AI Survey Instructions

Published on October 29, 2026 v1.19.0 We have further enhanced the problem-driven investigation entry point: investigations can now be initiated from issues submitted by engineers or from cluster cards, and we have improved the presentation of conclusions and the citation of evidence in Pod metrics charts. This is a version update, not the project’s initial release.Release Notes

How to Get Started

1. Prepare a kubeconfig that can access the target cluster. First, select the test environment and the minimum necessary permissions.
2. Depending on your operating system, choose the CLI or desktop version from the official installation documentation. On macOS/Linux, you can install it using Homebrew:brew install skyhook-io/tap/radar.
3. (of a computer) run kubectl radar, open the workspace in your browser; the default service address is localhost:9280.
4. First, review the resources, topology, and events, and then configure the supported AI CLI or MCP client as needed.
5. Before using AI to investigate, read the data stream prompts, review the evidence, and then decide whether to proceed with your own change process.Local Installation Guide

Costs and Deployment Options

The open-source version is licensed under the Apache 2.0 license; the software can be run for free on-premises or in a cluster without requiring a Radar account. The Cloud Free plan includes up to 3 clusters, unlimited users, and 100 hosted surveys per month; Team plans start at $149 per cluster per month, Enterprise plans at $299 per cluster per month, and custom plans are also available. Paid survey quotas, retention periods, and governance features vary by plan; the free quota for Cloud does not mean that accounts with built-in models are also free.Official Price

How to Choose Among Similar Tools

Basis for Comparison:K9s Official Website,Headlamp Official Website. This comparison focuses on how the methods work, not on a ranking of their effectiveness based on standardized testing.

Who Is It For, and What Limits Should Be Considered?

Radar is ideal for small engineering teams that frequently troubleshoot container application issues, as well as for platform teams seeking a unified view across multiple clusters. Its value lies in reducing the effort required to gather context, rather than guaranteeing that engineers identify the root cause correctly.

Running locally does not mean that all AI data remains on-premises: when a local investigation is enabled, the relevant read operations enter the processing workflow of the selected model provider. Investigations and applying fixes are separate steps; modifications made during supported on-premises workflows require confirmation, and cloud-hosted investigations do not provide an “Apply” operation that writes directly to the cluster. Different CLIs offer varying levels of isolation; “read-only investigations” should not be treated as a security guarantee for the entire on-premises environment. Production environments should still maintain access controls, change approval processes, and rollback mechanisms.Surveys and Data Boundaries

Official Website: Radar - Open Source Project