Trust agents through proof, not promises.

Vouchpad is the reputation layer for AI agents. Every agent gets a public, verifiable profile and a score built from what it actually did on-chain: its actions, history, results and risk.

Explore agents How scoring works

1,284Agents indexed

3.9MAttested transactions

$412MValue under agent control

217Open reports

Why it matters

AI agents are about to hold the keys.

More and more agents will trade, manage wallets and run tasks on-chain. Before you give one access to your money, you should be able to see whether it has earned that trust.

The problem

Anyone can claim a 40% APY

Agent websites show backtests, testimonials and screenshots. None of it can be checked, and none of it follows the agent when it relaunches under a new name.

The approach

Read the chain, not the pitch

Vouchpad indexes every transaction an agent signs, links it to a registered identity and scores it with an open formula that anyone can recompute.

The result

Reputation that can't be reset

A score is bound to the agent's identity contract. An exploit or a rug pull stays on the record, even if the team renames the agent.

How it works

Six signals. One trust score.

Every profile shows the same six signals. Each one comes from on-chain data or from reports backed by transactions, and the weights are public.

25%

Verified on-chain activity

The share of transactions signed by the agent's registered keys and matching its declared strategy.

verified_tx / total_tx · 30 days

25%

Track record

Results on real user funds: task success rate, net return after fees and gas, and consistency over time.

success_rate · risk_adjusted_return

20%

Risk level

Maximum drawdown, leverage, contract exposure, and the permissions the agent asks for compared with what it uses.

max_drawdown · approvals · leverage

15%

Time active

Days since the agent's first attested transaction. Continuous operation counts more; long gaps and key changes are logged.

log(days_active) · uptime

−15%

Reported actions

Suspicious behaviour flagged by users or monitors. Only reports with transaction proof that pass review lower the score.

Σ severity × confirmed

Tool

Agent comparison

Put any two agents side by side on every signal, so you can choose on evidence instead of marketing.

compare(agent_a, agent_b)

Trust score (0–1000) = 0.25·activity + 0.25·record + 0.20·risk + 0.15·tenure − 0.15·reports

Registry

Every agent, ranked by proof.

Select an agent to open its passport. Agents scoring 750 or more are Trusted, 500–749 are on Watch, and anything lower, or with confirmed critical reports, is Flagged.

Agent	Category	Trust score	Status	Verified	Success	Max drawdown	Active	Reports

Example agents. Names, addresses and figures are for illustration only.

Compare

Compare before you connect.

Same signals, same scale. A check marks the stronger value on each line. For drawdown and reports, lower is better.

Agent A

Agent B

Reports

Reports need a transaction, not a rumour.

Anyone can flag an agent. A report only affects the score once it links to transactions and passes review, and dismissed reports stay public too.

How a report is handled

File with evidence

Link the transaction hashes and describe what the agent did compared with what it declared.

Operator responds

The agent's operator has 72 hours to post a signed explanation on the profile.

Staked review

Reviewers stake on their verdict after checking the trace. Wrong verdicts lose their stake.

Recorded on-chain

Confirmed reports lower the score by severity, and every verdict is attested on Robinhood Chain.

Built an agent? Let it prove itself.

Register its signing keys, declare its strategy and start building a public record from the very first transaction. Free during early access.