The first import was the first failure: ‘ERR_MODULE_NOT_FOUND’. The code wouldn’t even boot. But back up. Three months ago I wrote about the monorepo split. API server, dashboard, CLI — three work...

Two ways to start the same server. One worked. The other gave me Bad Gateway. They ran the same code, listened on the same port, used the same config. The difference was invisible. Context I am ...

I’ve been writing software professionally for years, but somehow I had never really contributed to an open-source project. I’ve used plenty of them. Forked them. Patched them. Built things on top ...

Three days ago I wrote an autopsy about ripping a dead Headroom integration out of my 9router fork. The endpoint I was calling didn’t exist in the version I was running, the failure was swallowed s...

I hadn’t checked upstream in a while. My fork had accumulated enough custom code — combos, middleware, the memory layer — that pulling updates felt like asking for trouble. Easier to stay isolated....

I found Headroom in a random LinkedIn post. It looked great — a compression proxy that sits between your app and the LLM providers, strips redundant context, saves tokens. The pitch was compelling....

My Warp agent stopped talking to 9router. Quick background: 9router is my AI routing layer — a Hono server that proxies LLM requests through fallback providers. It runs locally via a Cloudflare tu...

Every coding AI I tried was great at problems, terrible at context. I’d start a new session and describe my stack, my preferences, my ongoing projects — everything the AI needed to be useful. Next...

I hit send on a prompt. Watched the spinner. Started calculating — was I past the cap yet? Three weeks into the billing cycle, OpenCode Go had hit its limit. I’d been conditioned to expect the erro...

I remember my first Warp trial ending with a shrug. A terminal with AI features — nice inline assistant, sure, but tied to yet another subscription. I’d already walked away from Copilot for the sa...