OpenUniverse

AI operations, grounded in production

OpenUniverse is my self-hosted AI operations platform. Agents investigate incidents and tickets, then act on servers through SSH, Ansible and OpenTofu. It is in private alpha and deployed at UNEP-WCMC with the organisation's permission.

What I built

As sole developer, I direct AI coding agents, set the design and test every change. The platform has a Go backend, Vue 3 frontend and 1,000+ automated tests.

• Provider-agnostic agents support Anthropic, OpenAI, OpenRouter and self-hosted models.
• An MCP server exposes 82 infrastructure tools; every call is logged with secrets redacted.
• Allow/ask/deny guardrails put human approval in Slack before actions that require it.
• Prompt caching cut input-token cost by approximately 90%.

Why it matters

This brings my production operations experience into agent engineering: useful infrastructure access, auditable actions and an operator in control. Built in my own time since February 2026.