An AI SRE that learns your logs, then pages you

What it does

Versus is a Go-based incident hub that tries to remove the busywork of writing alert rules. Its AI SRE agent—still in beta—reads your logs, builds a pattern catalog, and auto-creates incidents when it spots something it has never seen before. For everything else, it ingests webhooks from Alertmanager, Grafana, Sentry, CloudWatch, and similar tools, then fans out templated notifications to Slack, Microsoft Teams, Telegram, Lark, and others. Unacknowledged incidents escalate to AWS Incident Manager or PagerDuty.

The interesting bit

The agent uses a pattern miner with configurable similarity thresholds and tree depth to cluster log lines, and it offers training, shadow, and detect modes so you can preview its judgement before it starts waking people up. That graduated rollout is a thoughtful hedge against the usual “deploy and pray” anomaly detection.

Key highlights

• AI agent learns log patterns and auto-opens incidents for novel anomalies, with manual regex pre-filters to control noise.
• Webhook endpoint accepts alerts from any POST-capable tool and formats them with user-controlled Go templates.
• Notifications to Slack, Teams, Telegram, Viber, Email, and Lark from a single configuration.
• Embedded admin dashboard to review learned patterns and incident history.
• On-call escalation to AWS Incident Manager or PagerDuty when incidents are not acknowledged.

Caveats

• The AI agent is in beta, and its catalog storage currently supports only local file mode; redis and database backends are unimplemented config stubs.
• Redis is required for the agent to remember log source cursors across restarts, so persistent log tailing needs more than just disk.

Verdict

Teams tired of maintaining brittle alert rules or duct-taping webhooks to chat channels should try it; if your incident pipeline is already heavily customized, Versus may feel like a lateral move.