Autonomous Node Guardian

L1 & L2 Node 인프라 자율 운영

Geth, Reth, OP Stack, Arbitrum — 모든 EVM 노드를 AI가 24/7 자율 운영. 장애 감지, 정책 기반 플래닝, 승인 게이트 자동화로 팀의 운영 부담을 줄입니다.

terminal

$ cp .env.local.sample .env.local

$ docker compose up -d

✓ sentinai started on port 3002

✓ agent loop active — observing L2 metrics

지원 클라이언트

모든 주요 EVM 실행·합의·L2 클라이언트와 호환

L1 실행
GethRethNethermindBesu
L2
OP StackArbitrum NitroZK Stack
L1 합의
LighthousePrysmTeku

What it does

Five capabilities working together to reduce MTTR and operator burden for L2 infrastructure teams.

Real-time Detection

Monitors L2 operational anomalies in real time — sync failures, sequencer stalls, infra instability.

Policy-based Planning

Generates recovery plans based on risk tiers. Dangerous actions are blocked by default.

Auto-execution

Low-risk remediation actions (restart, scale) execute automatically within policy bounds.

Approval Gating

High-risk operations (drain, replace, rollback) require human approval via ChatOps.

Audit Trails

Every decision, action, and outcome is logged. Full traceability for ops teams and governance.

L1 검증자 모니터링

finality 지연, 피어 격리, sync 상태를 실시간 감지. L1 합의 레이어 이상을 L2 운영 전에 차단합니다.

How it works

A simple three-phase loop that runs every 30 seconds on your infrastructure.

01

Observe

Continuously ingests L1/L2 metrics, RPC responses, pod health, and logs from the chain infrastructure.

02

Decide

AI analyzes anomalies, scores severity, selects remediation playbook, and assigns a risk level.

03

Act

Executes low-risk actions automatically. Routes high-risk actions to on-call approval before acting.

Deployment Options

Start locally in minutes. Scale to production when ready.

Docker Compose

Local development and demo. Up in under 10 minutes.

$ docker compose up -d

  • Next.js dashboard on :3002
  • Redis state store
  • Caddy HTTPS proxy (optional)

Kubernetes (EKS)

Production-grade deployment with real K8s scaling integration.

$ AWS_CLUSTER_NAME=my-cluster docker compose up -d

  • Auto-detects EKS region & API endpoint
  • Real pod scaling (kubectl integration)
  • L1 RPC failover with auto-switch

Safety & Control

Autonomous doesn't mean uncontrolled. Every action is governed by policy, risk score, and human approval gates.

Risk-tiered Policies

Actions are classified by risk level. Each tier has its own execution policy — from auto-approve to always-block.

Destructive Actions Blocked

Node deletion, data wipes, and other irreversible operations are blocked by default at the policy layer.

Approval Required

High-risk remediations trigger a ChatOps approval flow. No action is taken without on-call confirmation.

Full Audit History

Every decision, action, and outcome is stored with timestamps. Replay past incidents for post-mortems.