Siclaw — AI Agent Platform for SRE

2 min read Original article ↗

Open Source — Star us on GitHub

AI Agent Platformfor SRE

Specialized AI agents for SRE. Deep investigation by default — read-only, open source.

Multi-AgentRead-Only by DefaultHypothesis-DrivenLearns From Every Incident

How It Works

Investigate, learn, repeat

A continuous loop that gets smarter with every incident

Multi-agent

One workspace, many specialists

Each agent owns a domain — Kubernetes, networking, system. Its own scope, skills, knowledge, and memory. They collaborate in Deep Investigation mode to solve issues that span layers.

k8s-agent

Kubernetes domain

Investigates pods, deployments, and cluster events.

network-agent

Networking domain

Traces latency, packet loss, DNS and routing issues.

system-agent

System / OS domain

Inspects CPU, memory, disk, and kernel-level faults.

Isolated by default, collaborative on demand. A pod restart in k8s-agent can hand off to network-agent for latency checks, then to system-agent for kernel signals — one investigation, three specialists.

Capabilities

Built for production

Multi-Agent Workspace

Specialized agents per domain. Each has its own scope, skills, knowledge, and memory.

Skill System

Per-agent diagnostic scripts with mandatory review. Fork and share across teams.

Knowledge Library

Per-agent wiki of runbooks and docs. Each specialist reads only what its domain needs.

Security First

Read-only investigation by default, with controlled execution when needed.

Alert-Driven Channels

Connect Siclaw to team channels and trigger investigations from alerts or operator input.

Cron Patrols

Schedule health checks in natural language. "Check GPU every 6h" just works.

Integrations

Connects to your stack

Built-in channels and Kubernetes. Extend to anything via Model Context Protocol.

Messaging

SlackDiscordTelegramLark

Observability

Prometheusvia MCPGrafanavia MCPElasticsearchvia MCPLokivia MCP

Alerting

PagerDutyvia MCPAlertmanagervia MCP

Dev

GitHubvia MCPGitLabvia MCP