An AI SRE that is actually on call
OpenSRE is an AI site-reliability agent. It watches the systems you point it at, triages what fires, digs through logs and metrics for the likely cause, and writes up what it found so a human can decide fast.
On call only counts if someone is awake. On Sokko, OpenSRE runs on its own always-on machine, so the 3am alert gets investigated at 3:01 and you read a summary instead of a wall of raw pages.
What OpenSRE does
Triages alerts before you wake up
When something fires, OpenSRE starts investigating immediately: what changed, what correlates, what looks like the cause.
Reads the boring data for you
Logs, metrics and recent changes get combed and summarized. You get the three lines that matter, not three thousand.
Writes incident notes as it goes
Every investigation ends in a write-up: symptoms, timeline, likely cause, suggested next step. Post-mortems start half-finished.
Keeps credentials contained
Read-only tokens for your monitoring and infra go in the Secrets tab and reach the agent securely, never a shared doc or a dotfile.
Deploy OpenSRE in three steps
Pick a plan
Plans start at Cookie, $12 a month, for the always-on machine. Model usage comes from your own API key or Sokko credits.
Create the agent
OpenSRE deploys from the CLI today: npm i -g sokko, then sokko instances create --runtime opensre. It comes up on its own always-on machine in about a minute, private to your team.
Connect your stack
Add read-only credentials for your monitoring, logs and alert channels in the Secrets tab, then tell OpenSRE what to watch and where to report.
What people run on it
First responder for a small team
Teams without a follow-the-sun rotation give OpenSRE the night shift. Humans wake up for confirmed problems, not for every blip.
Investigation sidekick during incidents
While humans coordinate, OpenSRE pulls timelines, checks recent deploys and drafts the running incident doc.
A weekly reliability report
It watches trends between incidents: error budgets, slow burns, flaky jobs, and files a short weekly summary before they page anyone.
Questions people ask
How much does OpenSRE hosting cost?
Plans start at $12 a month (Cookie) for one always-on agent, with model usage on your own key or Sokko credits. Cheaper than one hour of 3am engineering time.
Does OpenSRE change my infrastructure?
Give it read-only access and it investigates and reports without touching anything. How much access it gets is your call; start read-only.
Where do my monitoring credentials live?
In the Secrets tab. They are kept in secure storage and reach your agent directly, out of your code and out of chat logs.
Do I need to keep a server running for it?
No. The point of an AI SRE is that it is awake when you are not, so Sokko runs it on an always-on machine that restarts itself if anything crashes.