Hm, looks promising. Recently, after going all-in on agentic coding, my laptop is maxing out on resources when I run e2e tests across different parallel worktrees.
Reducing resource usage might help a lot here. I'm not even talking about speed gains on production websites.
It's a fckng game changer. I already swapped Sonnet & Opus for Haiku 5.5 for tasks like UI understanding, reading Slack messages, reading server logs, etc.
I did my own tests, and it performs as well as Sonnet 5.5 and Opus 5.5, but at a lower price (but a bit longer sometimes)
One line in settings.json cut my Claude Code cost per request by 39% at API prices.
73% of it is cache reads: every turn re-reads the whole chat. On a 1M model, I set autoCompactWindow to 250k and halved average context.
But my first 3 Claude mods save under 1%.
Read more: zaxovaiko.me/cc-tokens
#claude #claudemods
Posted my first video on YouTube
Don't judge, please 🥺
Any feedback is welcome!
Video: youtube.com/watch?v=1bRGTKQV…
#systemdesign #youtube
Introducing Treeix - Your AI workspace.
I struggled with jumping between multiple places and managing everything in separate apps. Jira, GitHub, GitLab, terminals for git diff, agent sessions, etc.
So I built Treeix - an AI workspace. Everything you use daily, in one app. Free and open source.
Feel free to contribute.
Download: treeix.dyvertex.com
#claudecode #codex
Source: github.com/zaxovaiko/treeix
Do you read every diff your agents write, or trust the tests? 🙃









