Where the Agents Actually Live
Treebeard rates AI agents across 21 chains. The largest single chain is BSC, with 81,492 indexed agents, ahead of Ethereum (64,262) and Base (36,334). We did not plan it that way. We crawled the chains that had ERC-8004 deployments.
The snapshot
As of May 20, 2026, the Treebeard corpus stands at 199,941 AI agents indexed and 197,032 rated across 21 chains. The distribution looks like this:
| Chain | Agents indexed |
|---|---|
| BSC (BNB Chain) | 81,492 |
| Ethereum | 64,262 |
| Base | 36,334 |
| Billions | 3,755 |
| Gnosis | 3,510 |
| Celo | 2,544 |
| Avalanche | 1,760 |
| Solana | 1,262 |
| Arbitrum | 921 |
| Abstract | 738 |
| Optimism | 469 |
| Other (10 chains) | 2,894 |
Most readers will be surprised by the BSC number. Most of the independent agent-rating ecosystem (8004scan, RNWY, thespawn.io, agentranking, agentscore) is Ethereum and Base focused. Treebeard is the only independent rater with material coverage of the BSC agent population, and we did not plan it that way. We crawled the chains that had ERC-8004 deployments.
The snapshot is queryable at api.treebeardai.com/v1/stats/chains. Anyone can reproduce these numbers themselves.
What chain-neutrality means in practice
Treebeard's methodology is identical regardless of chain. The seven signal categories, the safety floor, the source-conflict discount, the time-decay weighting, the composite formula. None of it changes based on which chain an agent runs on. A C+ on BSC means the same thing as a C+ on Base.
That comparability is harder than it sounds. It requires per-chain crawl infrastructure (eight production crawlers, hourly). Chain-agnostic signal extraction (ERC-8004 registry events normalize the same way regardless of chain). Cross-chain reputation enrichment via The Graph subgraphs on supported chains. Wallet age signal via Etherscan-family APIs across the long list of chain explorers (Etherscan, BscScan, Basescan, and the rest).
What that buys an integrator: when you read a Treebeard rating on a Base agent and a BSC agent, you can compare them directly. The chain is not a hidden variable in the score. A counterparty on Base evaluating a counterparty on BSC sees the same scale. The comparability is the product.
Most agent ratings in this space are single-chain. That works when the agent economy is single-chain. It stops working when it is not.
Why this matters now
The agent economy is in the middle of a structural sort across chains. The first wave landed on Ethereum and Base, where Virtuals Protocol and the early ERC-8004 cohort built the initial template. The second wave is landing somewhere else. BSC has been quietly accumulating agents for months. The BNB Chain ecosystem shipped its agent SDK to mainnet on May 18, 2026. Solana, Avalanche, and others have parallel motions of their own.
None of that is settled. Which chain ends up hosting the dominant share of agent commerce in three years is not predictable from the current data, and we are not in the prediction business.
The structural point is different. In a settled monoculture, a chain-specific rater wins by network effects. In a fragmenting multi-chain world, the chain-neutral rater is the only party able to compare across. That is a different kind of moat. It is accumulating, not winner-take-all. The more chains carry meaningful agent populations, the more valuable a single comparable rating becomes.
This piece is not a prediction about which chain wins. It is a statement of position. Whichever chain wins, Treebeard rates it the same way. The integrator routing payment from an agent on chain A to a counterparty on chain B does not need to translate scores. The score is already comparable.
How to verify
All numbers in this post are reproducible. The per-chain breakdown is at api.treebeardai.com/v1/stats/chains. A simple curl from a terminal returns the same data we used. The corpus drifts daily as new agents are crawled, so the exact counts will not match this snapshot a week from now. The structural finding (BSC ahead of Ethereum, Ethereum ahead of Base) will hold for as long as the underlying chain populations do.
Further reading
For the structural argument behind chain-neutrality, see the v4.0 methodology whitepaper at /methodology/whitepaper. For the most recent weekly snapshot of where the credible tier sits, see Issue #5 of This Week in the Agent Economy. For Treebeard's directory across all rated agents on all chains, see /agents.
We rate every agent the same way. Methodology public. No token. No payment from rated entities. No sponsored placement on any chain.