How evidence is collected
Every "used by" on this site links to a source you can open. 28,831 usages across the tools we cover; 17,415 of them from makers' live products, the rest from public code; 1,526 confirmed by two different kinds of source.
Five kinds of source, strongest first
The company names the tool in its own legal disclosures — privacy policy, subprocessor list, DPA, trust center. It answers for these, so it's the strongest proof a tool is in use now.
Limit: Only companies that publish them — mostly beyond the solo stage.
The tool's script on the product's pages, its DNS record (a CNAME to the vendor, an email provider's DKIM key) or its server header. Objective, and re-checked on every run.
Limit: Only what a browser can see: frontend scripts and infrastructure, not backend libraries.
A maker names the tool for their product — on its Product Hunt page ("built with") or in a founder interview on Starter Story — with the words quoted.
Limit: A snapshot in time: they may have switched since. Product Hunt shout-outs lean positive.
The tool's own site tells a named customer's story and what they use it for.
Limit: Real customers, but marketing, and dated.
A dependency in a public repository's manifest (package.json, requirements, go.mod…) or a config file (a workflow, a compose image). Verifiable by anyone.
Limit: Declared isn't necessarily used in production, and most repos are open-source projects rather than makers' products — so the site counts these second.
What the numbers mean
- Makers' products first. A usage counts as a maker's when any of its evidence comes from the live product — not only from a repo. Shares on decision pages ("what makers chose") use these alone.
- Observed, not market share. Hosted services show up on websites and in legal pages; libraries mostly in code. Each count says what could be seen, not how many people use a tool.
- Usage for the decision. A usage counts for the decision it serves: PostHog's analytics users aren't counted as feature-flag evidence.
- Never ranked. Tools are described by the situation they fit; counts are evidence, not a score.
Other data on the site
- Prices are read from each vendor's own pricing page, with the date checked, and modelled as plan fee + usage beyond what's included × the vendor's rate. Volume discounts, annual billing, taxes and regional prices are noted, not modelled.
- Adoption trends are monthly downloads of the tool's main package on npm or PyPI and GitHub stars — shown relative to the median package, since installs are up across the board.
- AI-agent readiness is checked from public sources: an llms.txt on the tool's site, an official server in the MCP registry, an official CLI on npm, and AI app builders' own integration docs.
- Which comparisons get a page follows what makers search: search autocomplete for every tool and decision.