SREs
Menu
Open workspace

AI-assisted incident response, with people in control.

Evidence before action.

SREs is an evidence-first incident response platform where specialized agents analyze observability data, identify likely root causes, and propose remediations that require human approval before execution.

Inspect the observations behind each finding.

Open the incident lab See the supporting evidence

This lab injects faults into bundled sample services and requires a working Groq connection. Every suggested fix still needs human approval.

A quiet mineral landscape framing an incident evidence excerpt
Inside an investigation

From observations to a decision

Start a run to collect evidence from the lab.

Workflow guide
What the agents doResults appear in your investigation
  1. Read measured request rates and durationsPrometheus
  2. Inspect recorded requests and errorsLoki
  3. Compare observations using original timestampsEvent analysis
  4. Evaluate causes and explain evidence gapsGroq
Human review requiredApprove a proposed lab action before execution

From warning to informed action.

SREs gathers the clues behind a service problem, explains the most likely cause, and recommends what a person can safely do next.

See an investigation in action
  1. 1

    Start a lab incident

    Choose a controlled fault. Agents inspect the resulting requests, metrics, and logs.

  2. 2

    Investigate

    Bring the clues together to explain what changed and identify the most likely cause.

  3. 3

    Approve

    Recommend a safe response, then wait for a person to approve or reject it.

See the evidence behind every recommendation.

Each finding stays linked to the data that supports it, so people can understand why the system suggested a cause instead of trusting a hidden AI decision.

A person makes the final decision
  1. 01
    Measured signals

    The lab sends real HTTP requests. Their responses and durations produce the logs and metrics agents inspect.

  2. 02
    Model analysis

    Groq interprets observations and proposes a supported lab action. Missing evidence can leave an investigation inconclusive.

  3. 03
    Reviewable records

    Open an investigation to inspect model inputs, tool responses, findings, and the human decision.

  4. 04
    Recovery checks

    After an approved control change, SREs tests application request paths and records whether they succeed within the lab latency limit.

Your investigation recordInspect the model and its evidence.Model failures stop analysis. Approved lab actions start the observed Redis container, terminate an observed database blocker, or roll back an observed release.
Browse actual investigations