Claude
AI Models
The majority of my daily AI work, and increasingly the thing that does the work rather than advises on it. Long context is why I haven't switched.
$20/mo
The complete directory behind the stack reports — what's running in production, what's on trial, and what got cut. No tool is here because someone paid for it to be.
Looking for machines rather than software? The local AI hardware table compares every unified-memory box and GPU on price, bandwidth and measured tokens per second — re-verified monthly.
The majority of my daily AI work, and increasingly the thing that does the work rather than advises on it. Long context is why I haven't switched.
Local inference for the unglamorous jobs — classification, extraction, template filling. Keeps API costs down and data on my desk.
The GUI answer to Ollama's CLI. Shows you exactly what will fit in memory before you download it, which is worth more than it sounds.
Saturates the benchmarks that used to separate models. Waiting on tier access before I can say whether the computer-use numbers survive contact with real work.
The backbone. Tens of thousands of executions a month on the Mac mini across lead gen, distribution, monitoring and agents. Nothing has given me a reason to move.
My agent framework, execute-first with a model fallback chain and a Telegram kill switch that has saved me from more than one runaway loop.
The plumbing that connects an assistant to your real tools. Genuinely useful, genuinely a new attack surface — pick servers the way you'd pick npm packages.
The alert and control layer for everything else. Free, instant, and the fastest way to give an automation an off switch you can reach from your pocket.
Still the best tech purchase I've made in years. Runs the entire self-hosted stack around the clock at roughly 15 watts.
128GB of unified memory for well under the NVIDIA ask. The value pick for local inference today — and a real PC the rest of the time.
Buys you prompt-processing speed and a frictionless CUDA stack. Doesn't buy you meaningfully faster token generation, which is what most people are actually shopping for.
The video pipeline: frames out of ComfyUI, Ken Burns and layered audio out of FFmpeg, long-form ambient video out the other end. Free, local, irreplaceable.
Graduated from trial to core this year. Takes 0% of subscription revenue, gives you a real website with real SEO control, and the free tier carried me a long way.
This site. Static output, component reuse, and clean URLs without shipping a framework to the browser. The rebuild took a weekend and paid for itself immediately.
Git push, site deploys, global CDN, $0. Replaced paid shared hosting entirely and made publishing a two-minute operation instead of a chore.
Where the sites live, and the trigger for every deploy. A commit is a publish — that single change removed more friction than any tool I bought this year.
The data half of the outbound engine — contact sourcing and enrichment. Paired with Instantly it runs largely unattended.
The sending half — inbox rotation, warmup, sequencing. The part that keeps cold email from becoming a deliverability disaster.
Named click events on everything that matters. Not because the dashboard is pleasant, but because guessing what converts is how you waste a year.
Genuinely excellent editor — wrong shape for how I ended up working. My work moved to delegate-and-review, which Claude Code covers inside a subscription I already pay for.
Nothing wrong with it — static sites on a global CDN with git deploys are simply a different category. Kept as the email provider for my domains, where it does fine.
This directory tracks what's in the monthly stack reports. Nothing appears here because a vendor asked — if a link ever earns a commission, it's disclosed on the disclosure page, and it never changes a verdict. Think something's missing? Tell me.