🤖

DevOps AI Bot — Diagnostic Architecture Waterfall

Zero-Token Priority: Local DB ➔ K8sGPT ➔ Local Log Engine ➔ Gemini Reasoning
← Back to Studio
DevOps AI Bot Diagnostic Architecture Flowchart

📊 Sequential Execution Order & Token Economics

Order Diagnostic Phase Sub-System / Engine Token Cost Latency What It Resolves
0 Healthy Workload Short-Circuit In-Cluster API Inspector 0 Tokens <15ms Running pods with 0 restarts and 0 warnings.
0.15 Sibling Deduplication In-Memory Workload Cache (15m TTL) 0 Tokens <1ms Pod storms (multiple replicas of same deployment/cronjob crashing).
1️⃣ Local Institutional DB PostgreSQL Vector / Trigram Memory 0 Tokens <25ms Errors already diagnosed and verified by human SREs.
2️⃣ K8sGPT CNCF Analyzer K8sGPT Custom CRDs & CLI 0 Tokens <30ms Deterministic infra issues: OOM Exit 137, ImagePull, Taints, Mounts. (Generic echoes ignored).
1.5 Confirm-Before-Escalate Gate Candidate Match Evaluator 0 Tokens <30ms Unverified candidates (60%-84% match) awaiting SRE confirmation before LLM spend.
3️⃣ Local SRE Log Engine Offline Universal Leaf Parser 0 Tokens <20ms UnknownHostException / DNS, Liquibase locks, missing ConfigMap placeholders.
4️⃣a Gemini + Prior Validation Google Gemini Flash-Lite (Tier 2) ~420 Tokens ~1.8s Validates and adapts an existing candidate fix when deeper analysis is requested. (75% token savings).
4️⃣b Gemini Cold SRE Reasoning Google Gemini Flash-Lite (Tier 3) ~650 Tokens ~2.5s Brand-new unknown errors. Fed with pre-computed clues, 84% cropped logs, and live dependency ground truth.