Topic
6 posts tagged “agents” on the ion blog: write-ups and how-tos from building ion, newest first. All posts
OpenClaw is a chat-first agent that lives in your messaging apps. Claude Code and Codex are coding agents that also run everyday tasks. A fair comparison of where each runs, billing, security and memory.
A concrete setup for an always-on Claude Code machine: power settings, tmux or launchd, Tailscale instead of open ports, where the token lives, usage limits and a queue of work.
Claude Code's push, ntfy hooks and the phone apps can all buzz your wrist, but few let you answer there. What reaches an Apple Watch today, what doesn't, and the options that do.
Lovable and Bolt, Claude Code and Codex, and Dyad compared: where each runs, who owns the code, and what it costs on top of a Claude or ChatGPT plan.
An honest beginner walkthrough: what you need, how to write a one-page brief the agent can build from, why to answer its questions first, how to check the result, and where first-timers get stuck.
My task runner limits concurrent agent runs with a single global number, not a quota per project. That makes a queued project look stalled, so here are the four signals that tell saturation apart from a real failure.
Invite-only, released in small waves. Free during early access.
