I built an AI observability platform with $0 – zero dependencies, zero ops, stateless

A developer built Cognilumin, a stateless AI observability platform, on $0/month infrastructure using free tiers of Gemini, Groq, and OpenRouter, with zero dependencies and no frameworks. The platform offers three tools: Illuminator for structured explanations, Topology Visualizer for animated system maps, and Execution Logs for tracing AI agent reasoning steps.

I’ve been building Cognilumin on and off for the past few months, mostly late nights and weekends. It’s a stateless AI observability platform—three tools that help explain, visualize, and debug AI behavior. The catch? It runs on $0/month infrastructure with no frameworks and no paid APIs. Here’s what it does: · Illuminator – paste any AI‑related question and get back a structured, master‑teacher explanation with analogies, mental models, and a key takeaway. The backend detects the intent evaluation, debugging, observability, or general teaching and selects a specialist prompt. · Topology Visualizer – describe a system scenario in plain English and a live particle‑based flow‑field engine renders an animated state map. Think lightweight Galileo Smartscape. · Execution Logs – give an AI agent a task and trace every reasoning step with hex‑formatted indices, action tags, and latency profiling. Browser‑based, zero‑setup LLM debugging. How I kept the stack at $0 · Hosting : shared hosting moved from InfinityFree to Hostinger for cron jobs · AI providers : Gemini, Groq, and OpenRouter—all free tiers, with a PHP proxy that rotates keys and handles 429s automatically · Storage : none—the platform is completely stateless no logins, no database, no tracking · Dependencies : zero—vanilla PHP + vanilla JavaScript, not a single framework What I learned along the way · Rate‑limit juggling is a real skill. Writing the fallback proxy Gemini → Groq → OpenRouter took a few hours, and it hasn’t failed once in production. What this is and isn’t This isn’t a SaaS. It’s a fully functional demo platform that proves AI observability can be done without expensive tooling. If you’re evaluating LLM pipelines, debugging agent behavior, or just want a lightweight way to visualize system states, Cognilumin works right now. 🔗 Live demo: https://cognilumin.com https://cognilumin.com I’m genuinely looking for feedback—what would you add, what feels clunky, what would make you actually use this regularly? If someone sees a bigger vision for the domain + codebase, I’m open to that conversation too, but honestly I’d just love to hear what the dev community thinks.