Investigate production incidents with a purpose-built SRE agent

A SaaS SRE team that struggled with general agents in incident response built its own Rust agent.

Difficulty
AdvancedOur rating
Time to build
Not reportedOur estimate
Monthly cost
Not reported
Revenue
Not reportedWe only show revenue the source states
Engagement
28 HN points6 comments · fetched Oct 15, 2026

What they did

jvogt’s team at Mezmo tried Claude, OpenClaw and LangChain in incident workflows but hit context overflow, security risks, hallucinations and frontier tokens wasted on easy work. They built Aura.

How it went

They open-sourced Aura, which investigates and fixes production incidents.

Worth knowing

They drew a hard line at relaxing production permissions; approval fatigue was real.

Builder recipe

Our own starting point, written from the source. Adapt it to your setup.

Stack

  • Aura
  • Observability data
  • Scoped credentials

Steps

  1. Write down the manual process step by step.
  2. Automate the deterministic parts with plain workflow steps or scripts.
  3. Use the model only where judgment is needed, with confidence checks.
  4. Add alerts and a human fallback.

Starter prompt

Incident: [alert]. Gather the last 30 minutes of related logs and deploys, rank three hypotheses with evidence, and propose the safest mitigation. Do not change production.

Open the repo ↗ (opens in new tab)

Tags

Tool Rust
Role SRE
Industry Software

Read the original

Related use cases

The Daily 5.

Five real AI agent use cases each morning, each with a recipe and a source. Plus a Sunday roundup. Free.