Skip to main content
OpenSRE

An agentic SRE that investigates production incidents autonomously — so your on-call rotation doesn’t have to start cold.

Point it at the tools your team already uses — Datadog, Grafana, Slack, PagerDuty, and 60+ others. It pulls the logs and metrics that matter, figures out what happened, and hands you a plain-English report.

​
Get started

Install OpenSRE

One command on macOS or Linux, then a quick setup wizard.

Run onboarding

Tell it which LLM and tools you’re using — opensre onboard walks you through it.

Connect your tools

Datadog, Grafana, PagerDuty, Slack, Kubernetes, and more.

Try an investigation

Paste in an alert, or just describe what’s going on.

​
What is OpenSRE?

OpenSRE is an open-source agent that investigates production incidents for you. When an alert comes in — or you just tell it something’s wrong — it goes and checks: recent deploys, logs, metrics, whatever’s connected. Then it writes up what it found. Who’s it for? Anyone on call who’s tired of starting from a blank tab at 3am. It’s built for engineers who want a head start on root cause, not a black box that hides how it got there. Why people like it:
  • It works with what you already have — no need to rip out your existing stack
  • You can run it locally, in your terminal, or hook it into a hosted setup for your whole team
  • It’s open source, so you can see exactly what it’s doing and change it if you need to
What do you need to get going? An API key for an LLM (or a local model), access to a couple of your tools, and about five minutes. The Quickstart walks through it.

​
What you can do with it

Investigate an alert

Get a written report with the likely cause, the evidence, and next steps.

Chat with it

Describe an incident in plain language, or use commands like /investigate and /verify.

Connect 60+ tools

Monitoring, incident management, cloud, code, chat, databases — most of what you already run.

Keep sensitive data out

Secrets and personal data get redacted before anything leaves your environment.

​
Try it now

1

Install

One command and you’re done — no dependencies to wrangle first.
2

Set things up

This launches a short wizard. Pick an LLM provider (OpenAI, Anthropic, a local model — whatever you use), paste in a key, and connect any tools you already have, like Datadog, Grafana, or Slack. It tests your key on the spot, so you find out about a typo now, not mid-investigation.
3

Give it something to investigate

Point it at a sample alert to see the whole thing run end to end:
Or skip the file and just talk to it — open the shell with no arguments and describe what’s going on in your own words:
Once it’s done, you’ll have three files worth a look: problem.md (what it was looking at), the theory/hypothesis_*.md files (what it tried), and report.md (what it concluded). Prefer JSON? Add --output ./rca.json.
Something not working? Quickstart troubleshooting covers the usual suspects — Docker not running, make missing, LLM not configured.

​
How an investigation runs

Say an alert fires. Here’s roughly what happens next:
  1. OpenSRE reads the alert and figures out what kind of problem it’s looking at
  2. It gathers context — who owns the service, what changed recently, what’s normal
  3. It goes and checks the tools you’ve connected for logs, metrics, and traces
  4. It keeps digging and testing ideas until it’s confident it found the cause
  5. It writes up a report and sends it wherever you asked — Slack, a file, your API
Investigation workflow
How OpenSRE Works
Want more detail? How an investigation works walks through each step.

​
Keep exploring

Showcase

See real output from sample investigations.

Features

A rundown of everything OpenSRE can do today.

Shell command

Every slash command the interactive shell understands.

FAQ

Quick answers to common setup and usage questions.