Launch HN: Sonarly (YC W26) – AI agent to triage and fix your production alerts

Source: Hacker News

Overview

We’re building Sonarly (https://sonarly.com), an AI engineer for production that connects to observability tools such as Sentry, Datadog, or user‑feedback channels, triages issues, and automatically generates fixes to dramatically cut resolution time.

The Problem

• Production teams receive dozens to hundreds of alerts each day (e.g., 180 alerts/day from Sentry).
• With small on‑call teams, filtering noise, finding the real signal, and reproducing the root cause consumes a huge amount of time.
• Bugs in production can lead to churn or frustrated users, and they arise from code errors, database mismatches, infrastructure overload, or user‑specific behavior—issues that are hard to catch with E2E tests or AI code reviews alone.

Our Solution

Sonarly reduces MTTR (Mean Time To Repair) by:

1. Eliminating noise – grouping duplicate alerts and filtering false positives.
2. Providing plain‑English RCA – the AI explains the what, why, and how to fix each issue.
3. Integrating, not replacing – we connect to existing tools (Sentry, Datadog, Slack, etc.) instead of requiring a new tracker.

How It Works

• Data ingestion – We use the open‑source Sentry SDK (backend) and a custom frontend tracker built on rrweb. Adding a DSN to an existing Sentry config sends data to Sonarly without changing the monitoring stack.
• Signal extraction – Alerts are deduplicated and prioritized by impact on users or infrastructure.
• Contextual AI coding – Once a clear signal is identified, we invoke Claude Code with the exact context (Sentry issue, relevant logs fetched via grep on Datadog/Grafana, etc.).
• Dynamic system map – An internally maintained Markdown file maps services, logs, and metrics, giving Claude Code a quick overview of the production architecture, even in multi‑repo, multi‑service environments.

Example Workflow

A user receiving ~180 Sentry alerts per day followed this process:

1. Receive the alert.
2. Either get distracted from the current task or ignore the alert.
3. Open dashboards to locate the root cause (infra vs. code).
4. Determine if it’s a false positive, a known issue, or a new problem.
5. Provide Claude Code with the appropriate context to generate a fix.

After implementing Sonarly’s noise‑reduction layer, the same team saw:

• Alerts reduced from 180 → 50 per day by grouping duplicates.
• Severity scoring based on user/infra impact, yielding ≈5 actionable issues each day.

Triage now consists of three automated steps:

1. Deduplication before invoking the coding agent.
2. Root‑cause gathering for each distinct alert.
3. Re‑grouping by RCA to present a concise list to engineers.

Results

• Noise reduction: 70%‑80% fewer alerts presented to on‑call engineers.
• MTTR improvement: Faster identification and automated fixing of issues cuts mean repair time dramatically.

Demo

Watch a short walkthrough: https://www.youtube.com/watch?v=rr3VHv0eRdw

Call for Feedback

We’ve launched a self‑serve version (https://sonarly.com) with a generous free tier. We’d love to hear from engineers about:

• Your current alert‑handling workflow (Sentry, Datadog, user feedback, etc.).
• How you assign ownership and gather context for fixes.
• Any automated processes you use to filter noise or remediate bugs.

You can sign up and try Sonarly in under two minutes. My co‑founder and I will be in the thread to answer questions, share technical details, and gather constructive feedback.

---

Comments URL: https://news.ycombinator.com/item?id=47049776