I built a daily-verified radar for free AI APIs and GPU credits

Free AI tiers and GPU credits change faster than a bookmark list can keep up.

I kept running into the same problem: a provider page would still be bookmarked, but the free quota had changed, a card had become required, or the offer was no longer available in my region. So I built AI Resource Radar — an open-source, local-first tracker that checks allow-listed public sources and keeps the evidence attached to every result.

The radar focuses on the decisions developers actually need to make:

The public site is read-only and needs no account or API key. The same data and dashboard can also run locally.

A long list of links is easy to publish but hard to trust months later. AI Resource Radar uses source-specific parsers and records source health separately from offer status.

If one parser breaks, the radar fails closed: it keeps the last trusted observation and marks the source as partial instead of silently deleting the offer. Severe integrity problems stop a new public snapshot from replacing the previous healthy one.

Community directories can help discover candidates, but they cannot promote an offer to “officially verified.” That requires evidence from an official provider source.

The latest release handled a real example of why this approach matters. Hugging Face's official pricing table changed the included monthly Inference Providers credit for Free Users from $0.10 to None.

The radar did not remove the old record after a single page observation. It retained the trusted value, required repeated successful confirmation, and then retired the obsolete offer. The release also made scenario publishing safer: a page is withheld when there is not enough verified evidence, rather than publishing a thin or misleading result.

At the time of this launch, the public manifest reports 23/23 official sources healthy and 186 resources in the current snapshot.

Install it in an isolated Python 3.11+ environment:

python -m pip install ai-resource-radar ai-resource-radar start --open Enter fullscreen mode Exit fullscreen mode Or run it without a permanent install:

uvx ai-resource-radar start --open Enter fullscreen mode Exit fullscreen mode The project is MIT licensed. I would especially value feedback on two things:

If you find stale data, there is a correction path in the project and every public record links back to its evidence.

Disclosure: I maintain AI Resource Radar. This launch article was prepared with an AI coding assistant and reviewed against the current public manifest, repository documentation, and signed v0.9.1 release.