RunbookAI
userunbook.ai
- Free plan
- Linux
What RunbookAI is
RunbookAI is a self-hosted, open-source incident-response tool for SRE teams investigating production incidents. It gathers incident context, ranks hypotheses, tests them against infrastructure, identifies a root cause, and recommends remediation. Users can ask natural-language questions across AWS, Kubernetes, and CloudWatch, then execute step-by-step runbooks with approval gates for mutating actions. Queries are read-only by default, and every infrastructure change requires explicit human approval. RunbookAI records hypotheses, evidence, proposed actions, approval decisions, and execution results in a local audit trail. It can index knowledge from local files, Confluence, and Google Drive. Documented integrations include AWS, Kubernetes, PagerDuty, OpsGenie, Slack, and Claude Code. A self-hosted shared knowledge server exposes runbooks, postmortems, and known issues through REST and MCP endpoints; CLI mode is also available. The maker says it runs inside the user’s infrastructure, sends no telemetry, and sends LLM calls directly to the configured provider without proxying or storing prompts. It is released under the MIT License and has no hosted service or control plane.
Who it suits
RunbookAI suits SRE teams running production workloads on AWS and Kubernetes. It fits organizations wanting self-hosted incident investigation, approval controls, and local audit records.
Good to know, and what to watch
- Ranks and tests incident hypotheses
- Read-only queries by default
- Human approval for mutating actions
- Local audit trail
- MIT-licensed and self-hosted
- No hosted service or control plane
- Requires Bun 1.0+ or Node.js 20+
- Requires an Anthropic API key for Claude
The verdict
RunbookAI combines incident investigation, runbook execution, operational knowledge, and approval controls in a self-hosted tool. Its local logging and stated no-telemetry approach suit teams keeping operations inside their infrastructure.
What the maker publishes
- Audit trail
- RunbookAI logs investigation hypotheses, evidence, proposed actions, approval decisions and execution results locally. source
- Claude Code
- The shared knowledge server exposes an MCP endpoint that Claude Code can use to query operational knowledge. source
- Data handling
- The maker says RunbookAI runs within the user's infrastructure, sends no telemetry, and sends LLM calls directly to the configured provider without proxying or storing prompts. source
- Deployment
- RunbookAI offers CLI mode and a self-hosted shared knowledge server, with no hosted service or control plane. source
- Deployment and license
- RunbookAI is self-hosted and released under the MIT License. source
- Incident workflow
- Its investigation workflow gathers incident context, ranks hypotheses, tests them against infrastructure, identifies a root cause and suggests remediation. source
- Infrastructure queries
- Users can ask natural-language questions across AWS, Kubernetes and CloudWatch. source
- Infrastructure queries
- The CLI supports natural-language queries across AWS, Kubernetes, and a knowledge base. source
- Integrations
- Documented integrations include AWS, Kubernetes, PagerDuty, OpsGenie, Slack and Claude Code. source
- Intended users
- The maker describes it as an incident-response tool for SRE teams and is onboarding teams running production workloads on AWS and Kubernetes. source
- Investigation
- Its investigation workflow gathers incident context, tests ranked hypotheses against infrastructure, and suggests remediation steps. source
- Knowledge sources
- RunbookAI can index knowledge from local files, Confluence and Google Drive. source
- Open source
- RunbookAI is released under the MIT License. source
- Purpose
- RunbookAI investigates production incidents by forming hypotheses, gathering evidence and recommending fixes. source
- Requirements
- The documentation lists Bun 1.0+ or Node.js 20+ and an Anthropic API key for Claude as requirements. source
- Runbooks
- It can execute step-by-step runbooks, with approval gates for mutating actions. source
- Safety
- Infrastructure queries are read-only by default, and every mutating action requires explicit human approval. source
- Safety controls
- Every mutating action requires explicit human approval, while infrastructure queries are read-only by default. source
- Shared knowledge server
- A self-hosted server lets a team query shared runbooks, postmortems and known issues through a REST API and an MCP endpoint. source
- Support
- The documentation directs users with questions or issues to GitHub Issues or Discussions. source
Checked against the maker's pages on 4 Oct 2026.