I build tools for working with AI agents, and I use them all day to find out which ones hold up. Almost everything here started the same way: something in the daily loop broke or annoyed me, so I built the thing that fixed it.
Voice AI infrastructure at @VapiAI during the day.
- temple — an IDE for your Claude Code and Codex agents. Sessions organized by project, all a click away.
- folio — a read-optimized Markdown app for macOS. Agents write most of my docs now; this is the reader for them.
- recall — hybrid semantic + full-text search over every Claude Code and Codex session you've run. Nothing leaves the machine.
- tarp — a plain, modern terminal. Warp with the AI, the cloud, and the account system pulled out.
- abra — hold Fn, speak, release. On-device dictation for macOS.
- raven — terminal dashboard for the Claude Code sessions running on your machine, showing which one needs you.
Plus smaller ones: later (leave a reminder for a
future agent session), routines (recurring
chores that don't get missed just because the machine was asleep),
ccmeter (what's left of
the subscription, and of the context window),
clipwire (paste screenshots into a CLI running
over SSH).
A theme, if there is one: local-first, files on disk, no telemetry, no account.