Keep Hits 12k Stars Unifying 130 Monitoring Tools With AI Alert Correlation
Keep unifies alerts from 130+ monitoring tools into one dashboard, adds LLM-driven correlation, and automates response with YAML workflows.
Takeaways
- Keep is an open-source AIOps platform aggregating alerts from 130+ monitoring tools into one dashboard (GitHub).
- YAML workflows work like GitHub Actions for monitoring, chaining triggers, filters, and actions across Slack, Jira, and more.
- Pluggable LLM backends (Anthropic, OpenAI, Ollama, DeepSeek, Gemini) power alert correlation and summarization.
- CEL-based rules engine plus indefinite alert history enables temporal correlation across recurring incident patterns.
- MIT core is free to self-host; managed cloud runs from free Startup tier up to $199/mo Growth and custom Enterprise.
- Deploys on Docker, Kubernetes, AWS ECS, or OpenShift with SSO, SAML, OIDC, LDAP, RBAC and ABAC.
Alert fatigue is the quiet productivity killer of every on-call rotation. Keep, a Python-based open-source project that recently crossed 12k GitHub stars, tries to fix it by consolidating alerts from every monitoring tool you already run and layering AI-driven correlation and workflow automation on top.
The platform unifies more than 130 monitoring tools into a single pane of glass, aggregating signals from providers like Datadog, Prometheus, Grafana, CloudWatch, New Relic, and PagerDuty so teams stop switching between dashboards to figure out what broke. The architecture centers on three components: a provider system for bidirectional integrations, a correlation engine powered by pluggable LLMs, and a workflow automation system that feels like GitHub Actions for monitoring infrastructure.

Cutting through the noise
Most incident tools centralize notifications without recognizing that a dozen alerts describe the same underlying database outage. Keep's approach leans on a rules engine plus LLMs to group related events. A Google CEL-based rules engine evaluates incoming alerts against user-defined correlation rules, automatically bundling related events into high-fidelity incidents. One useful architectural detail: Keep stores alert history indefinitely by default (configurable via retention policies), which enables temporal correlation. In practice, that means recognizing patterns like "this alert always precedes that alert by five minutes" and proactively grouping them, something most incident management tools can't do because they don't retain sufficient history.