Sherlocks is not a Rootly replacement. It solves a different problem.
Rootly coordinates incidents. Sherlocks investigates them. The two sit next to each other: Rootly pages the right people and opens the channel; Sherlocks posts the root cause with evidence into that channel within minutes of the alert firing.
The product is a Watson agent coordinating 16 specialist sub-agents (Kubernetes, databases, networking, certificates, cost, and more), deployed inside the customer's cluster with a service account that has no write verbs in its RBAC. Investigation is in-cluster. Writes are impossible unless the team explicitly moves specific actions up the autonomy ladder.
Teams running Rootly for coordination that want a real investigation agent on top, and teams in regulated or security-sensitive environments where read-only enforcement has to be provable.
- Four levels of autonomy you earn. Level 1 (read-only, system-enforced via Kubernetes RBAC) is the default. Level 2 (one-click human approval), Level 3 (allowlisted automatic actions), and Level 4 (denylisted) are configurations you turn on per-action, not products you buy.
- System-enforced read-only. Two read-only Data Agents run inside the cluster. Credentials and logs stay inside the customer's boundary. The service account cannot write, no matter what the agent decides.
- Watson plus 16 sub-agents. Each investigation surface has a dedicated agent. A single incident routes to the ones that matter, in parallel, and returns a cited root cause.
- Proof points. Fynd cut MTTR 70%, saved 77% of infrastructure engineering hours, and reduced support hours 63% across 800+ engineers (Fynd case study).
- Write-back to existing tooling. Root causes are posted to the ticket in the customer's own ITSM (ServiceNow, Jira, and others), so engineers never leave their workflow.
- Sherlocks does not replace PagerDuty or Rootly for paging or status pages. If your bottleneck is paging rotations, this is not the right tool.
- Kubernetes-centric. Non-Kubernetes estates (bare EC2, Lambda-only, serverless-only) have less coverage today.
- In-cluster deployment takes longer than SaaS integrations. First investigation usually runs within a week; full rollout to every environment takes 4 to 8 weeks.
- Pricing
- Free tier of 30 investigations per month. Enterprise is custom pricing with unlimited investigations.
- When to pick it
- Keep Rootly for coordination and add Sherlocks for investigation. Or replace Rootly with incident.io for coordination and run Sherlocks next to it. Either combination is valid; Sherlocks does not try to be both halves of the product.