🛡️ Aether-Guard V2
Autonomous AI SRE Agent with Production-Grade Safety
What is Aether-Guard?
Aether-Guard is an autonomous AI SRE agent that detects incidents, performs root-cause analysis, and executes automated remediation actions with production-grade safety guarantees.
Key Features
- Hybrid RCA Engine: 60% deterministic rules (<50ms, free) + 40% LLM-powered analysis (2-5s, Claude Sonnet 4.5)
- 6-Layer Safety Pipeline: Rules → Policy → Approval → Remediation → Verification → Auto-Rollback
- Production Ready: 280 passing tests, full CI/CD, observability with Prometheus/Grafana/Tempo
- Cost Efficient: 60% cost reduction vs pure LLM approach ($2.40/month for 1000 incidents)
- Works Offline: Deterministic rule layer continues functioning even when Claude API is unavailable
Architecture Highlights
| Component | Technology | Purpose |
|---|---|---|
| Agent | Python + Claude 4.5 | Hybrid RCA, policy enforcement, verification |
| Listener | Python FastAPI | Alertmanager webhook receiver, enrichment |
| Target Service | Go 1.25 | Demo workload with chaos injection endpoints |
| Storage | Postgres + Redis | Incident analytics, RAG similarity search |
| Observability | Prometheus + Grafana + Tempo | Metrics, dashboards, distributed tracing |
Quick Start
Clone the repository and start the full stack with Docker Compose:
git clone https://github.com/jnzm02/Aether-guard.git
cd Aether-guard
cp .env.example .env
# Add your ANTHROPIC_API_KEY to .env
make docker-upThis starts:
- Target service (port 8080) - demo workload with chaos endpoints
- Listener (port 8081) - webhook receiver
- Agent (port 8082) - AI SRE brain
- Prometheus (port 9090) - metrics
- Grafana (port 3001) - dashboards
- Postgres + Redis - persistence
Project Status
Note: This is an active research project demonstrating production-grade AI agent safety patterns. The V2 architecture is complete with full test coverage and CI/CD. Ongoing work focuses on reliability improvements and expanded pattern coverage.
Current Version: V2 (Phase A-C complete)
- ✅ Phase A: Production-grade target service with 9 validated chaos patterns
- ✅ Phase B: Hybrid RCA engine (deterministic rules + LLM fallback)
- ✅ Phase C: RAG-augmented investigation with similarity search
- ✅ 280 tests passing (Go + Python)
- ✅ Full observability stack (metrics, logs, traces)
- ✅ Policy-based safety gates and auto-rollback