Agent SRE

An on-call engineer that never sleeps, and never acts without telling you.

It sweeps your monitoring on a schedule, opens and tracks incidents, applies the fixes your procedure allows, asks before anything risky, and files the root cause with a suggested fix.

# incidentsSlack
[INC-142] API latency Sev 2 mitigating
WHAT BROKEAPI p95 latency above 2 s since 02:09. WHAT I KNOWThe worker queue backed up after the 01:58 deploy. WHAT I DIDRestarted the worker. Your procedure allows this. WAITING ONThe alert to clear on the next sweep.
Approval needed: scale the database to the next size Blast radius: brief connection drop on the primary. Rollback: scale back down. Nobody else touches this incident until it is decided.

How it works

Tell it your stack and your rules. It takes the first shift.

01 · DESCRIBE

Your stack and your procedure

In plain language: what runs where, what it may do on its own, and what needs a person.

02 · CONNECT

What it watches and acts on

Grafana, AWS, Fly.io, or any monitoring tool with an MCP server.

03 · SWEEP

On a schedule

Every few minutes it checks alerts and dashboards. Quiet sweeps send nothing.

04 · TRACK

One incident per problem

Each alert gets a fingerprint, so there are no duplicates. Severity, status and a full timeline.

05 · ACT

Within your rules

Allowed fixes happen right away and are announced. Risky ones wait for approval.

06 · TELL

Like a good on-call hand-off

What broke, what it knows, what it did, what it is waiting on. Escalates when it is stuck.

07 · ROOT CAUSE

A fix-ready ticket

Once things are stable, it reads your code, finds the cause, and files a Linear issue with the evidence and a suggested fix.

Your rules, enforced

You write the procedure. It follows it.

✓Allowed actions happen on their own, but never silently ✓Risky actions raise a gate that states the blast radius and the rollback ✓It never silences an alert or edits your monitoring ✓An incident is resolved only when a sweep sees the alert clear ✓It never edits code. The fix goes to people, or to Agent SDLC

From alert to fix

With Agent SDLC, an alert can end as a merged fix.

Agent SRE files the fix-ready ticket. Assign it to Agent SDLC and the fix goes through design, build and review, then waits at your merge gate.

Alert fires Incident opened Mitigated within your rules Root cause from your code Linear issue + suggested fix Assign to Agent SDLC Your merge gate

Connects to

Watches your stack. Talks where your team is.

WATCHESGrafanaAWS CloudWatchFly.ioAny MCP server
TELLSSlackDiscordEmail
FILESLinear

Ask it anything

Is production OK? One open incident: [INC-142], mitigating. Everything else has been quiet since yesterday's sweep at 18:00. What happened last night?

Answers come only from the incident record, never from guesswork.

Put an agent on first response tonight.

We will connect your monitoring, write the first procedure with you, and tune it on your real alerts.

Talk to us