OpsKit vs HolmesGPT
HolmesGPT is an Apache-2.0 CNCF Sandbox agent that investigates incidents across 50+ observability sources. OpsKit lives inside your coding agent with an enforced read-only cluster connection. An honest comparison.
Facts about HolmesGPT taken from github.com/robusta-dev/holmesgpt, checked 29 July 2026. Anything their site doesn't state is marked "not stated" rather than guessed at.
What HolmesGPT is
An open-source (Apache 2.0) AI agent for investigating production incidents, integrating Prometheus, Grafana, Datadog, Kubernetes and 50+ other data sources. A CNCF Sandbox project from Robusta.
What OpsKit is
A Kubernetes and cloud investigation kit for AI coding agents. It adds a read-only MCP connection to your live cluster, investigation skills with explicit evidence requirements, and a safety layer that blocks every mutation verb before it executes.
The short version
HolmesGPT is the more capable incident investigator on breadth of data sources — if your observability stack is wide and you want an agent that can reach all of it, it is hard to beat, and it is free. OpsKit is narrower on purpose: it runs inside the coding agent you already use, needs no extra service, and enforces read-only access at the OS level rather than by configuration. Choose on whether you want a dedicated investigation service or a kit inside your editor.
Side by side
| Dimension | OpsKit | HolmesGPT |
|---|---|---|
| Shape of the tool | A kit installed into your AI coding agent | A standalone agent you self-host, with an optional operator mode |
| Data sources | Kubernetes via MCP, plus AWS, GCP, Azure and Terraform agents | 50+ — Prometheus, Grafana, Datadog, Loki, Elasticsearch, cloud providers |
| Can it change things | No — mutations are blocked by an allowlist hook before execution | Read-only by design and RBAC-respecting; Operator Mode can apply fixes |
| Opens pull requests | Yes — resolves the owning repo from ArgoCD or Flux annotations | Yes — can open PRs when connected to GitHub |
| Runs where | In your terminal, inside Claude Code, Cursor, Codex CLI or Kiro | Self-hosted; Operator Mode runs in Kubernetes |
| Licence and price | Commercial — $19 one-time | Apache 2.0 — free, CNCF Sandbox |
Choose HolmesGPT if…
- Your investigations need to span Prometheus, Grafana, Datadog, Loki and cloud APIs in one agent.
- You want a service that reacts to alerts on its own, not a tool you invoke from a terminal.
- You need to investigate VMs, databases and SaaS platforms as well as Kubernetes.
- Open-source licensing and CNCF governance matter to your procurement.
Choose OpsKit if…
- You want zero extra infrastructure — no service to deploy, operate, or get budget approval for.
- You want the guarantee to be enforcement, not configuration: mutations blocked by a hook you can audit.
- You want the investigation in the same window as the code you're about to change.
- You want the redacted tool output bundled onto the PR so reviewers replay exactly what the agent saw.
Questions
Is OpsKit an alternative to an observability platform?
+
No. OpsKit reads the Kubernetes API and, where present, your GitOps state. It does not replace Prometheus, Grafana or Datadog, and it will say so explicitly when a question needs metrics it cannot see rather than guessing at the answer.
Can HolmesGPT and OpsKit run against the same cluster?
+
Yes. Both read through the Kubernetes API and respect RBAC, so read-only access for each is independent and safe to grant.
Try the free tool first
The Kubernetes YAML linter runs the same class of checks OpsKit's deployment guard applies — free, in your browser, no signup. It's the cheapest way to see how the kit thinks.