Jev-driven SRE diagnosis
SREGym blog post on Kubernetes incident diagnosis with no LLM agent: Jev picks a root cause from collected evidence.
This summary is based on how curated lists describe the project; its README was not available to review.
Overview
This SREGym blog post describes diagnosing Kubernetes incidents without an LLM agent. Code collects evidence from the cluster and Jev picks a root cause and its supporting observations.
According to the list summary, the approach passed 80 of 105 diagnoses across 21 fault types at a median of 14.6 seconds. Further details, such as how evidence is gathered and how options are framed, were not collected here.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which root cause explains the observed Kubernetes evidence?(inferred) | — |
| Yes / no | Does this observation support the chosen root cause?(inferred) | — |
How it works
- Collect Kubernetes evidence for an incident
- Present candidate root causes to Jev
- Jev selects a root cause and supporting observations
- Score diagnoses against known injected faults
Strengths
- No LLM agent loop required
- Reported median diagnosis time of 14.6 seconds
Caveats
- Only a summary was collected
- 80 of 105 diagnoses passed, so some faults were missed
At a glance
- Type
- Article
- Status
- research
Use cases and integrations
How curated lists describe it
Incident diagnosis: runs with no LLM agent at all, collecting Kubernetes evidence for Jev to pick a root cause and its supporting observations, and passes 80 of 105 diagnoses across 21 faults at a median of 14.6 seconds.
yibie/awesome-jev
Frequently asked questions
What does Jev decide?
The root cause of a Kubernetes incident and which observations support it.
How well did it do?
The summary reports 80 of 105 diagnoses passed across 21 faults.
Is an LLM agent involved?
No, the post describes running without one.
Related projects
On-call tooling where Jev judges alerts and incident hypotheses and plain code decides: alert routing plus a no-LLM root-cause benchmark.
ProjectbetaPythonUnix filter for security alert triage: JSONL alerts in, Jev answers five typed questions, code routes each alert.
ProjectbetaRustIncident-scoping tool that starts from one malicious process and asks Jev which linked telemetry belongs to the same incident.
Demoprototypegrep by meaning: pipe text in, ask a yes/no question in plain English, and Jev returns matching lines, even behind tail -f.
BenchmarkbetaPythonGo CLI for managing WordPress sites at scale, with typesafe commands that ask Jev questions and a scan --triage that ranks malware findings.
ProjectproductionGoExperiment testing whether a local Qwen read via label log-probabilities can replace Jev on synthetic AIOps alert decisions.
ProjectresearchPython