Where to Start

Last updated: March 23, 2026

This is more infrastructure than most people need. Each piece is simple on its own: skills are markdown, the daemon watches files and runs scripts, mesh is JSON files in a directory, Q reads and writes markdown. The composition is complex. I've been iterating on this for months.

You don't need all of it to start getting value.

Start with session history. That's the minimum viable version of everything above. Every AI coding session produces a conversation log. Most people throw these away. Keep them. Summarize them, search them, extract patterns, feed them back into future sessions. You can do this tomorrow with any agent.

Encode what you learn. Next time you solve something tricky, write it down where your agent can find it. A skill, a runbook, an AGENTS.md section. The time saved applies to every future session. Build a meta-skill and the bootstrapping compounds from there.

Try one extra agent. Spawn a second agent in a tmux window for a focused task: run tests, explore a codebase, review a PR. See if the workflow clicks before investing in coordination infrastructure.

That's it. Everything else in these pages grew from those three seeds.

pi-mesh is MIT licensed. Bosun is the harness. The ideas are transferable to any setup.