VersusControl/versus-incident

What it solves

Versus Incident is a self-hosted AI SRE agent designed to reduce alert fatigue by automatically detecting anomalies in logs without requiring manual alert rules. It prevents the noise of known patterns from reaching engineers, escalating only new or unexpected issues to chat channels and on-call platforms.

How it works

The system operates through two primary intake methods:

1. AI SRE Agent: This agent reads logs from sources like files or Elasticsearch. It uses a "training" mode to learn normal log patterns and a "detect" mode to identify brand-new errors or anomalies. When a new pattern is found, an AI SRE triages the event, generating a summary, severity level, and suggested next steps.
2. Webhook Alerts: It acts as a centralized hub for existing monitoring tools (e.g., Prometheus Alertmanager, Grafana, Sentry, CloudWatch SNS), receiving alerts via JSON POST requests and routing them through the same notification pipeline.

Once an incident is raised, it is formatted via Go templates and fanned out to multiple channels (Slack, Teams, Telegram, etc.) and escalated to on-call providers (PagerDuty, AWS Incident Manager) if not acknowledged in time.

Who it’s for

DevOps and Site Reliability Engineers (SREs) who want to automate log-based anomaly detection and centralize their incident notification and escalation workflows.

Highlights

• Zero-Rule Anomaly Detection: Automatically learns "normal" log patterns to trigger alerts only on new issues.
• Multi-Stage Deployment: Includes training, shadow, and detect modes to validate AI judgment before going live.
• Broad Integration: Supports a wide array of notification channels (Slack, Teams, Telegram, Viber, Email, Lark) and on-call platforms.
• Flexible Filtering: Uses regex pre-filters to drop boring noise (like 200-OK responses) before the AI processes the logs.