Foglamp is an observability platform for AI agents built on the Vercel AI SDK that surfaces the cost, latency, token usage, distributed traces, and output quality of every LLM call. A two-line SDK instruments every generateText and streamText call, then exposes per-agent spans, latency, and spend along with the full call flow. It scores production traffic with code checks and LLM judges (evals), waterfalls every run with the exact prompt and response per span, and fires threshold alerts on cost, latency, and error rate. Cost intelligence breaks down spend by model, agent, and customer so teams can catch regressions — like a 10x cost spike — before users complain.