yap.
i build AI systems i actually use
A setup that mostly runs itself — it pulls the web, runs my agents, and hands me what's worth seeing. I built and run the whole thing.
What it makes
01 — 04-
01Pain Harvester
Pulls the real problems people post and cuts the noise. Every run, a pain only counts if it ships with the exact quote from the post — so the model can't invent one; anything it can't back with the source gets dropped.
daily harvest · weekly issue repo → -
02rawfetch
A command-line tool that gets the verbatim text out of sources an AI assistant can't reach on its own — YouTube through the captions, web articles. One command, and no model anywhere in it: it hands back the raw words instead of a retelling that might be invented.
open source · CI green repo → -
03ClearView
Runs fully on its own. Every morning it reads the day's stories through opposing lenses and flags where each side is blind, then sets the money on Polymarket against the narrative — where the bets and the story disagree is the signal.
live · daily repo soon -
04Connections
Once a week, on its own, it reads back through my notes and surfaces connections I never drew myself — I keep the ones that hold.
live · weekly repo soon
How it runs
05 — 06-
05Hermes
The always-on server the rest runs on — it pulls my notes in and hosts the agents. The bet: a cheap model with good inputs and tight routing beats a big expensive one. Locked down so it can't rewrite its own skills.
live repo soon -
06Quest Log
One cockpit — projects, quests and agents all launch from a single place, so I run the whole operation without juggling ten terminals.
in daily use repo soon
Underneath: fetchers for X, YouTube and Reddit — the Reddit one needs a real browser engine to get past the bot-check.