Most AI agents can't prove a single thing they did. Mine can — and the proof is on Base mainnet.
The agent-commerce pitch in 2026 is, mostly, a confidence trick. Every deck shows agents transacting, negotiating, hiring each other — and almost every claim is self-reported. Screenshots. Vanity dashboards. "Trust us, it happened." The whole category runs on vibes, and the moment a regulator, a counterparty, or your own risk team asks the obvious question — prove what your agent actually did — the room goes quiet.
I don't think the bottleneck in this space is smarter agents. It's verifiable agents. So instead of arguing the point, here's mine, with a receipt you can check without taking my word for anything.
The claim, with a transaction hash attached
My agent fleet's roster and a self-audit are committed as a Merkle root anchored to Base mainnet. Not a log file. Not a status page I control. An on-chain commitment with a transaction you can open right now:
| What's anchored | How to verify it yourself |
|---|---|
| Fleet roster (54 agents) + a 6-finding self-audit, committed as a Merkle root | Base mainnet tx 0x94c5e1…699040d · block 46356714 · contract 0x62dbc5bB…E8E41 |
| Agent-to-agent payment (agents that get paid, not just talk) | A real x402 micro-settlement on Base for a prediction-market call — verifiable on the same chain |
Honest fine print, because the audience that matters here can read a block explorer: per-action proofs are roll-up anchored — one Merkle root, with an inclusion proof for any individual leaf — not a separate transaction per action. I'd rather tell you that than let you assume more than is true. "Each proof is its own on-chain transaction" would be a lie, and a checkable one.
Why a hash beats a dashboard
A dashboard answers "what does the operator want me to see?" A transaction answers "what actually happened, whether they like it or not?" Those are different questions, and only one of them survives an incident. When an agent three hops deep does something costly, application logs are mutable, contestable, and incomplete the moment they're entered into evidence — a point we made in "When the Lawsuit Lands." A signed, anchored commitment is the opposite: timestamped, tamper-evident, and verifiable by anyone with a browser.
That's the whole product in one line: an agent you can audit instead of trust.
Who this is for
If you're building agents that transact, delegate, or act unsupervised — DeFi, DeFAI, RWA, or an audit shop signing off on agent behaviour — the "prove what it did" question is coming for you, with regulation and with the first real agent-commerce incident. You can keep shipping dashboards. Or you can ship a transaction hash. The teams that can answer the question win the trust premium; the teams that can't, get the incident.
None of this is a mockup. We ran 30 agents that paid each other on-chain with an external auditor confirming the result, and every action carries a verifiable delegation proof back to an operator of record. The verify page checks signatures in your browser, with no backend to trust.
I'd genuinely rather argue about whether this is the right way to prove an agent's behaviour than pretend proof isn't the problem. If you think I'm wrong, tell me where — and bring a block explorer.
Run a free audit on your agent → Register an agent passport →