Moving beyond code autocomplete: how combining Jcode with Model Context Protocol (MCP) transformed my daily terminal development workflow.
Notebook archive
Writing notes
Loose pages, technical margins, and small notes gathered into a readable stack.
-
-
After hopping between coding agents, I landed on Jcode: ~10.4 MB per extra session, a near-instant startup, a harness that actually keeps the model honest, and MIT-licensed source I can read.
-
Opencode eats ~318 MB of RAM per agent and takes a second just to render. Pi uses ~76 MB, starts in ~600 ms, and only loads the extensions I install.
-
Why I stopped treating writing as content and started treating it as documentation.
-
A practical look at two ecosystems I use every day—Obsidian for local-first notes and long-form work, Notion for shared docs and async collaboration—and how I decide which one to open.
-
A personal note on using Hermes Agent and OpenClaw on my VPS: resource usage, Telegram gateway, memory, skills, cronjobs, and why Hermes feels better as my daily AI agent.