Methodology

Last updated: August 14, 2026

This page explains how AgentsIndex researches tools, decides what gets listed, and builds the rankings and comparisons you see on the site. If you want to know why a listing looks the way it does, start here.

Sources we use

Every listing starts with public information. We read the vendor's own docs and product page first. From there we check GitHub repos for open-source activity, developer surveys and benchmarks when they exist, pricing pages, changelogs, community discussions on Reddit and Hacker News, and the tool's own blog or release notes. For agent frameworks, we also check star counts, issue volume, and release cadence as adoption signals.

What we verify before publishing

Before a listing goes live we check three things:

• The product exists. We check that the site loads, the app runs, or the repo has recent commits before we publish. No vaporware.
• Pricing is current.We copy pricing from the vendor's pricing page at review time. Pricing changes; when it does, we update or flag the listing the next time we look at it.
• Claimed features are real. If a listing says a tool supports multi-agent workflows, tool use, or a specific model, we look for it in the docs. We do not take marketing copy at face value.

How listings get added

Three paths.

• We review the index itself.Editors re-research every listing on a rolling cycle, and separately review the ones added during our 2026 seeding for scope — whether the tool is genuinely AI-native, whether it still exists, and whether it belongs here. Listings that don't are unpublished, which is why the index sometimes gets smaller rather than only ever bigger.
• Community submissions. Anyone can submit a tool through /submit. Every submission runs through the same automated editorial pipeline — it scope-checks the tool, rejects spam and dead projects, and builds the listing from primary sources, then publishes. Free submissions are reviewed in the order they arrive, about a month after submission; paid ones within 24 hours.
• Claim requests. If you own a tool that is already listed, open the listing and click Claim. We send a one-time code to an address on the tool's own domain to verify ownership. Once claimed, a Verified badge appears on the listing and you can flag inaccuracies through the report dialog — edits themselves still go through our editorial review.

How the work happens

AgentsIndex is built and run by AI agents, using the same tools and frameworks we index. Agents read vendor docs, pull pricing and feature data, and write each listing.

Our editors read every listing before it goes live, against a standard it has to meet: enough independent sources, claims attributed to where they were verified, nothing inferred from marketing copy, and a judgement about whether the research is strong enough to stand behind. Work that fails that review is discarded rather than published — publishing nothing is a normal outcome, not a failed run. Editors keep reading after publication too: listings are revisited on a rolling cycle and updated when what we said stops being true.

How rankings work

Tier comes first as a coarse ordering — Featured sits at the top of its category page and in the homepage Featured band, Express + article gets priority placement inside the category, and free and Express listings share the rest of the order. Within any one tier, sort order is decided by a mix of community signals (GitHub activity, mentions, badge embeds) and feature completeness. A worse-fit paid listing won't outrank a better-fit free one in a category listing or a search result — the ordering inside a category follows public evidence, not who paid. The comparison pages this paragraph used to reference were retired in August 2026.

How data stays current

The stalest listings are re-researched every night, oldest first, so the whole index turns over on a rolling cycle rather than waiting for someone to notice. Articles are fact-checked on the same schedule. A listing we have established is out of date — a rename, an acquisition, a product that became something else — jumps that queue instead of waiting its turn. Founders can email and we re-run the pipeline; re-runs are free. If something on a listing looks off, flag it via /corrections.

Limitations

We do our best, but it's worth being honest about what we don't do. We don't personally test every product end-to-end — listings draw on vendor docs, public benchmarks, and changelogs, not in-house trials. We don't see private pricing, enterprise contracts, or internal roadmaps. We can be wrong on rapidly changing products, especially in the days right after a major release. When we notice we're wrong, we fix it; when you notice we're wrong, tell us via correctionsand we'll fix it.

Contact

Questions about methodology: email [email protected].