tokenbill
Estimated API costs for Claude Code sessions, broken down from local logs. No network calls.
Explore project ↗Engineering AI that matters.
AI engineer at Arena Labs.
I work on agents, retrieval systems, and the tools around them.
awesome-llm-on-anything
A curated list of language models running locally on retro PCs, consoles, and microcontrollers.
Explore project ↗homeostat
Policy constrained, verifiable self healing for dedicated Android devices.
Explore project ↗marketmind-ai
A multi agent platform for market analysis and sentiment.
Explore project ↗Latest writing
All articles →Claude Opus 4.8 and the Question Nobody Wants to Ask: Are Frontier Models Hitting a Plateau?
A look at whether progress in frontier models is slowing down.
Yapay Zekâ Kullanımında Kurumsal Sorumluluk ve Veri Gizliliği
İş yerinde yapay zekâ kullanımı, kurumsal sorumluluk ve veri gizliliği üzerine.
Projects.
My tools and projects, plus a collection of language models running on unusual hardware.
tokenbill
Understand where a Claude Code session’s estimated API cost goes, including token usage, cache behavior, and context compaction. Reports are generated from local logs.
Usage & report options
Run this from your project directory:
npx tokenbillExport a self contained HTML report:
npx tokenbill --htmlThese are API equivalent estimates, not your subscription bill.
awesome-llm-on-anything
A curated collection of language models running on unusual hardware, from retro computers to game consoles and microcontrollers. Entries link to the original projects and reported results.
homeostat
Policy constrained, verifiable self healing for dedicated Android devices.
marketmind-ai
A multi agent platform for market analysis and sentiment.
Writing.
Claude Opus 4.8 and the Question Nobody Wants to Ask: Are Frontier Models Hitting a Plateau?
A look at whether progress in frontier models is slowing down.
Yapay Zekâ Kullanımında Kurumsal Sorumluluk ve Veri Gizliliği
İş yerinde yapay zekâ kullanımı, kurumsal sorumluluk ve veri gizliliği üzerine.
LLM Inference: Insights into Memory and Performance
VRAM, KV cache, latency, and throughput: the practical pieces of LLM inference.
Implementing RSA for Digital Signature from Scratch
Building RSA key generation, signing, and verification in Python.
Learn.
01. Workflow or agent?
Choose between a fixed sequence and a model that decides the next step.
02. Prompt chaining
Break a task into small steps. Check each result before moving on.
Emrehan Gorgec.
I’m an AI engineer at Arena Labs. I work on agents, retrieval systems, and the tools around them.
Outside work, I build the projects collected here and write about what I’m learning.