About Infrabase

Infrabase is a directory of AI infrastructure: inference APIs, vector databases, agent frameworks, observability tools, the things you assemble around a model. A bit over 200 products, each one written up by hand.

That last part is the only thing that distinguishes it. Most AI directories are scraped, so they carry whatever the vendor's homepage claimed the day the crawler ran. This one is slower and smaller as a result, and the data is worth more.

I'm Arvid, a developer in Sweden. I run this on my own. Email me with a question, a correction, or a product worth adding.

How listings get made

A person reads every product before it appears here. Pricing, HQ country, GDPR status, licence and what the product actually does get checked against the vendor's own site, not against search results or another directory. Anything I can't confirm gets left out rather than guessed at.

In practice that means catching things like a provider whose cheapest listed price turned out to be its speech-to-text model rather than an LLM, or a repository whose licence was reported as permissive everywhere except in its own LICENSE file. Both were wrong here first, and both got fixed by opening the primary source.

It still goes stale. Prices change, companies get acquired, projects get archived. Every product page has an edit link, and corrections go into the source data. Submitting a product is free and always will be: submit one here.

No rankings, no "best" list

There is no "best" field here and no editorial ordering. Which tool wins depends on constraints only you know, and whatever is fastest or cheapest this quarter often isn't next quarter. The comparison pages put sourced details next to each other and stop there.

For agents and LLMs

The whole directory is queryable as JSON at /api/query. No key, no signup. Filter by category, job, GDPR status or HQ country, or search across names and descriptions. The endpoint docs list every parameter; agents.md covers which filter to use when, and where the data is weaker than it looks.

There is also a snapshot on GitHub as a plain README. Both are CC-BY-4.0: use them however you want, commercially included, as long as you credit Infrabase. If you are building on this data and need a field that isn't there, ask.

How this pays for itself

Two things cover the costs, and both are the kind of thing that quietly corrupts a directory, so here is exactly what each one buys.

Some outbound links are affiliate links and earn a commission if you sign up through them. Having one does not get a product listed, does not change how it is written up, and does not move it up a page. Most listings here have no affiliate programme behind them and you cannot tell which from reading them.

Companies can also buy a featured listing for $29 a month. That pins their product to the top of its category page with a label on it. It does not reorder anything else, and the API ignores it entirely. What nobody can buy is a mention in a comparison, a blog post or the Builder Radar.

Arvid
@arvid_a · hi@infrabase.ai