What they built

Cloud Guard AI

Enterprise cloud environments leak money and expose attack surface quietly. GPU nodes idle at 2% CPU while billing $3/hour. An S3 bucket sits publicly readable because someone loosened an ACL during debugging and never reverted it. Finding these means manually correlating inventory, utilization telemetry, security findings, and the org directory that tells you who to actually contact. CloudGuard AI turns that recurring audit into a conversation. It's an MCP server exposing 10 tools that let any LLM reason over live infrastructure data — correlating hourly cost against CPU utilization to compute exact monthly waste, auditing findings by severity, resolving resources to named owners with escalation paths, and generating Terraform remediation. It files real GitHub P1 tickets, posts Slack Block Kit incident cards with approve/escalate buttons, and closes issues once remediated. The agent has no write access to infrastructure — not by permission, but because those tools don't exist in the schema.