Production-Grade Engine v0.1.0 Released

Autonomous AIOps Incident Remediation Engine.

Zervox bridges raw Prometheus metrics, pod logs, and safe Kubernetes execution. Ingest alerts securely, diagnose with LLMs, enforce OPA policy guardrails, and execute remediation automatically or via human authorization.

curl -sSL https://raw.github.../install.sh | bash
Documentation โ†’
zervox copilot interactive simulation
Zervox โšก Engine Ready. Ingestion listening on /api/grafana_webhook [Bearer Auth Active]
Click one of the buttons above to test live incident correlation and policy guardrail execution...

Engineered for Production Safety

Zervox delivers low-latency incident response with strict security boundary guarantees.

๐Ÿ”

Authenticated Webhooks

Enforces Bearer Token / API Key validation on all HTTP endpoints. Unauthenticated alert payloads are immediately rejected with HTTP 401.

๐Ÿ›ก๏ธ

OPA Rego Policy Guardrails

Validates all AI recommendations prior to execution. Absolute blocks prevent namespace deletion, RBAC edits, secret reads, and container shell exec.

๐Ÿ™‹โ€โ™‚๏ธ

Human-in-the-Loop Gateway

Enable ZERVOX_REQUIRE_APPROVAL=true to transition non-read-only cluster actions to PENDING_APPROVAL in SQLite instead of auto-executing.

๐Ÿ—„๏ธ

Resilient State Storage

SQLite operating in WAL mode with 5-second busy timeout connection pools guarantees zero lock contention or crashed tasks during telemetry spikes.

Absolute Safety Boundaries

Never worry about autonomous destructive cluster changes.

The Zervox Policy Engine acts as an unbypassable firewall between LLM recommendations and your Kubernetes cluster. Even if an AI model hallucinates a dangerous command, Zervox enforces absolute blocks.

โœ•
Prohibited Namespace Deletion
Blocks kubectl delete namespace
โœ•
Prohibited Container Shell Exec
Blocks kubectl exec -it commands
โœ•
Prohibited RBAC Modifications
Blocks ClusterRole & RoleBinding edits
โœ•
Prohibited Secret Inspection
Blocks direct reading of K8s Secrets

System Architecture

Non-blocking, event-driven remediation pipeline built with Axum, Tokio, and kube-rs.

[Grafana / Prometheus Alert] โ”€โ”€(Bearer Auth HTTP POST)โ”€โ”€> [Axum Ingestion Server]
                                                             โ”‚
                                                             โ–ผ
                                              [(SQLite WAL DB Store)]
                                                             โ”‚
                                                             โ–ผ
                                              [LLM Anomaly Correlator]
                                                             โ”‚
                                                             โ–ผ
                                              [OPA Policy Engine Check]
                                              โ”Œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ดโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”
                                              โ”‚ (Approved) โ”‚ (Violation)
                                              โ–ผ โ–ผ
                                      [kube-rs K8s Client] [Block & Alert Team]
                                              โ”‚
                                              โ–ผ
                                      [Kubernetes Cluster Remediation]

Frequently Asked Questions

How does Zervox ensure LLMs don't delete cluster resources? โ–ผ

Zervox runs all LLM recommendations through an explicit OPA Policy Engine before executing any API call. Destructive operations like namespace deletions, RBAC edits, secret reads, and container shell exec are hard-blocked by policy.

Can I run Zervox with local LLM models like Ollama? โ–ผ

Yes! Zervox supports OpenAI-compatible API endpoints. You can point the LLM analyzer to a local Ollama instance running llama3 or mistral.

How does Human-in-the-Loop approval work? โ–ผ

When ZERVOX_REQUIRE_APPROVAL=true is set in your environment, any non-read-only action is stored with status PENDING_APPROVAL in the SQLite database and requires explicit admin authorization before execution.