SRE Intelligence Platform

An alert fires in Slack. The platform investigates it the way a senior engineer would, and replies in the same thread with what it found and what it recommends.

What happens when an alert fires

flowchart TD
    Alert["An alert posts into a<br/>Slack channel you subscribed"] --> Decide{"Is this worth<br/>investigating?"}
    Decide -->|No| Record["Recorded, and nothing else.<br/>No model call, no page"]
    Decide -->|Yes| Incident["An incident opens,<br/>attached to that thread"]
    Incident --> Gather["It gathers evidence from the<br/>tools you connected"]
    Gather --> Answer["It replies in the thread:<br/>what it found, ranked causes,<br/>the safest next step"]
    Answer --> Human["You reply, ask, correct,<br/>approve, or ignore"]
    Human --> Gather
    Answer --> Verify["When the alert clears it checks<br/>whether the problem is really gone"]

What it will not do

These are design decisions, not missing features.

• It never changes your infrastructure. Every connector is read-only. It holds no credential that can restart a pod, revert a deploy, or silence an alert. When it recommends one of those, a person does it.
• It never names a person as a cause.
• It never changes an incident's status because of something said in conversation. Status changes are explicit.
• It never claims a service recovered just because an alert stopped firing. It verifies first.
• It never falls back to a second model vendor. Silently answering from a different model with different behaviour is worse than saying the engine is unavailable.

Where to start

You want to	Go to
Run it on your own machine	Get started
Understand a screen you are looking at	Dashboard
Manage your workspace, its sign-in methods, and its members	Use
Give it access to your systems	Connect your tools
Know how it reaches its conclusions	How triage works
Configure the model, budgets, and access	Operate
Change the platform itself	Build

Two ways to work with it

In Slack. The investigation appears in the alert's own thread. Reply to ask questions, correct it, or approve a recommendation. You do not have to open anything.

On the dashboard. The same conversation, plus everything Slack deliberately leaves out: every lookup it made, the evidence behind each claim, the alternatives it ruled out, and what it cost.