How The Agents Index researches AI agents

The AI-agents space moves fast and markets itself hard. A directory is only useful if every entry is something you can trust — so here is exactly how we build and maintain each listing.

Every listing passes a quality gate

Before an agent can be published, its record must carry a minimum of structured, sourced facts — what it is, what category it belongs in, whether it is open-source, whether it exposes an API, and its pricing model. A record that cannot meet that bar simply does not publish. Thin, near-duplicate pages are structurally impossible here. The full bar is published on our methodology page.

We separate stable facts from fast-moving ones

Some facts are stable (what an agent does, whether it is open-source). Others move constantly (exact prices, model support). We record the stable facts plainly, describe pricing by its model rather than pinning numbers that go stale, and mark every listing with the date it was last verified.

We re-check on a schedule

Listings are re-researched over time, and changes land in the record — not silently. Each fact carries the source it was verified from — an agent’s license, for instance, is read from its own repository (like Aider’s), never from a marketing page — so you can check our work.

We grow deliberately, not in a flood

We add genuine, researched agents one at a time. You will never see this index dump hundreds of auto-generated pages online — real coverage, earned honestly, is the whole point. For the method applied in practice, see our report on terminal AI coding agents.