wtf is instruments?

Instruments is a small collection of two automated data tools plus shared plumbing, designed to run cheaply on a tiny Debian server with just 512 megabytes of memory and one CPU core. Both tools quietly gather data on a schedule and publish the results as simple websites, so no one has to remember to run anything by hand. The first tool, called the Observatory, tracks the cost of running AI models locally on your own hardware. It combines sold prices for used graphics cards scraped from the web with community submitted benchmark results, showing how many tokens per second of AI inference you get for every hundred dollars spent on a given card, so someone shopping for local AI hardware can see which cards offer the best value. Benchmarks come from running a tool called llama-bench and dropping the output into a folder, and numbers that look implausible are automatically flagged and left out of the rankings. The second tool, called the Registry, keeps track of which smart home devices still work if their maker's cloud service shuts down, alongside a running log of past cases where a company turning off its servers left customers' hardware unusable. Both tools are live on their own websites. The project follows a few deliberate design rules. Every scheduled job records whether it succeeded or failed, and only after three failures in a row does it send a single alert message through Telegram, rather than repeatedly nagging. If nothing is reported, everything is assumed healthy. The public site always keeps showing the last good data even if a scraping job is currently failing. All the gathered information builds up in a local SQLite database file, treated as the real long term asset, while the specific marketplaces and websites it scrapes are seen as replaceable sources. Getting started involves copying an example environment file, filling in an optional Telegram bot token and a free API key from a pricing data provider, installing a couple of lightweight Python dependencies, and running a handful of command line steps to initialize the database, load benchmarks and prices, and publish the site. A cron schedule can then automate the whole thing going forward.