Relvy AI

What is Relvy AI?

Relvy AI is a Y Combinator F24 startup founded by ex-Google, Uber, and Microsoft ML engineers that's tackling one of the most painful parts of software operations: on-call debugging. When a PagerDuty alert fires at 2am, Relvy AI investigates it automatically — connecting to your observability stack, querying relevant logs and metrics, and finding the root cause without requiring a human engineer to do the initial triage.

Key Features

Automated Root Cause Analysis — Relvy connects to your monitoring and logging infrastructure (Datadog, Grafana, CloudWatch, etc.) and automatically investigates alerts when they fire. In 70%+ of cases, it finds the root cause before the on-call engineer finishes opening their laptop.

Purpose-Built Debugging LM — Relvy ships a small language model specifically optimized for infrastructure debugging. Rather than using general-purpose LLMs (expensive, sometimes unfocused), their specialized model matches GPT-4o performance on debugging benchmarks at approximately 1/100th the cost per inference.

Runbook Execution — When root cause is identified, Relvy can execute pre-approved remediation steps from runbooks automatically — resolving common incidents without human intervention.

Alert Enrichment — Even for incidents that require human investigation, Relvy enriches the alert with gathered context: relevant log excerpts, metric graphs, recent deployments, similar past incidents. Engineers start debugging with a head start rather than from scratch.

On-Call Integration — Works within PagerDuty, OpsGenie, and Slack-based incident workflows. Investigation results appear in the incident channel automatically.

Why Relvy Matters

On-call burden is one of the largest hidden costs in engineering organizations. The $35B/year estimate in on-call costs includes not just the cost of the engineer's time, but the productivity tax from sleep disruption and the retention risk from chronic on-call stress. Automating the initial investigation phase reduces both.

Bottom Line

Relvy AI is early-stage but technically credible — the team's ML background and the specialization of their debugging model (rather than wrapping GPT-4) suggests genuine technical differentiation. Worth evaluating for teams with significant on-call load and monitoring infrastructure already in place.