Akmatori vs K8sGPT

Akmatori turns production alerts into AI-assisted incident response workflows for SRE, platform, and DevOps teams. Its agents collect logs, metrics, deployment context, and runbook knowledge to investigate incidents and suggest next actions. Alerts can come from Alertmanager, PagerDuty, Grafana, Datadog, and Zabbix, while Slack channels can also be monitored. Agents access infrastructure through SSH, Kubernetes, scripts, and API clients. Proposed risky actions require approval in the web interface or Slack before execution. Akmatori supports OpenAI, Anthropic, Google, OpenRouter, NVIDIA NIM, MiniMax, and OpenAI-compatible endpoints, including on-premises models. Recurring investigations can run as cron jobs with configured prompts and tool allowlists, posting results to Slack. Deployment uses Docker Compose or Helm and can be fully on-premises for air-gapped setups. The maker says incident data stays in the customer’s infrastructure unless an external provider is configured. The Apache 2.0 project provides a REST API with documented incident operations, authentication, webhooks, and OpenAPI 3.0.

K8sGPT scans Kubernetes clusters with SRE-grade analyzers and explains problems in plain English. It provides a CLI, an in-cluster operator, and an MCP interface for AI assistants. The project lists 23 built-in analyzers and supports custom analyzers over gRPC. Fourteen cloud or local AI backends are listed, including OpenAI, Azure OpenAI, Amazon Bedrock, Google Vertex AI, Ollama, and LocalAI. Documented integrations include Prometheus, Trivy, Kyverno, KEDA, and AWS. Analysis without the explain option can run without an AI backend, while explanations require one. Local or air-gapped operation can use Ollama, LocalAI, or a custom REST backend. Anonymization masks names and identifiers before prompts are sent. Optional operator auto-remediation uses a configurable risk threshold. CLI, Docker, and operator paths require a reachable cluster and kubeconfig; operator installation requires Helm 3.