flightcheck is an autonomous monitoring agent. It inspects a live LLM application through the Arize Phoenix observability platform, detects when output quality degrades, and files a real alert — no human in the loop.
On every check it sequences multiple tool calls — stats, traces, judgment, action — not a single canned reply.
A purpose-built Model Context Protocol server wraps the Arize Phoenix client and gives the agent its observability superpowers.
When it finds degraded quality, flightcheck files a severity-tagged alert to the team — autonomously.