avenox.lol   ·   YouTube / @avenoxai   ·   X / @Avenoxai

What I am building now

FLAGSHIP / IN ACTIVE DEVELOPMENT Serai The neutral control plane for every AI agent. AI agents are powerful, but the infrastructure around them is still broken: sessions start without the right memory, context stays trapped inside separate tools, risky actions lack consistent gates, and there is rarely a reliable record of what ran or why. Serai sits outside the model and gives Claude Code, Codex, Cursor, Hermes, and other agents a shared layer for scoped memory, compiled context, governed tool routing, human approvals, action receipts, workflows, and version history. It does not replace the agent or run the intelligence. It makes the same agent more reliable, inspectable, and able to carry work across sessions, tools, and teams. Serai is currently being built in private. It will be open-sourced in the future. serai.run   ·   live hub

Building what watches. I build AI systems, agent infrastructure, and strange little products in public. The polished explanations live on YouTube. The raw prompts, model outputs, failures, and working code live here. AI SYSTEMS  AGENT INFRA  MODEL TESTS  OPEN BUILDS

THE LEMON · quality control

Model tests

Same prompt. Different frontier models. One self-contained HTML file each.
Nothing quietly fixed after the model finished.

Test archive	What is inside
🍋 limon-arena	The growing archive of raw frontier-model tests from the videos
AI Ne Kadar İlerledi?	2023 → 2026 across four one-shot visualization tasks
Sonnet 5 vs Opus 4.8	Five identical prompts and both models' untouched output
GPT-5.6 SOL Benchmark	Four single-file demos with their original prompts

_{Everything from the videos lives here — raw, inspectable, exactly as it ran.}