Scouts
Sweep deployer wallets, token transfers, LP movements, contract events. Surface anomalies and feed candidate evidence into the pipeline.
SLEUTH AI runs adversarial due diligence on crypto projects. Scouts pull on-chain evidence, analysts model it, verifiers try to break it. The final report only ships after the dispute window closes — every claim linked to a transaction, contract event, or IPFS snapshot.

Four specialized agent classes operate in parallel. Scouts gather, analysts reason, verifiers attack, synthesizer commits.
Sweep deployer wallets, token transfers, LP movements, contract events. Surface anomalies and feed candidate evidence into the pipeline.
Build cash-flow, supply-distribution, and risk models from scout evidence. Score qualitative claims against quantitative on-chain reality.
Stake $SLTH against the report. Earn by breaking citations: dead links, fabricated tx hashes, or claims unsupported by the cited event.
Resolves contradictions, weights claims by verifier outcome, and emits a single signed report. Anchors a content hash to IPFS + Base.
Agents never trust each other. Every handoff is hashed, every claim is cited, and verifiers profit when something doesn't add up.
A user posts a target — contract address, deployer, or domain — paid via x402 USDC. Job is registered on Base and broadcast to the swarm.
Scout agents pull tx history, contract events, and IPFS snapshots in parallel. Each evidence node is hashed and pinned before downstream agents touch it.
Analysts build supply, liquidity, holder-concentration, and behavior models. Every claim is bound to one or more evidence hashes.
Staked verifiers race to falsify claims. Broken citation → slash on the original agent, bounty to the verifier. Dispute window stays open ~24h.
After the window, the synthesizer reconciles surviving claims into a single report. CID is anchored on-chain. Access gated by x402 microtxn.
Trust isn't assumed — it's purchased. Every citation has a bounty on it during the dispute window.
| id | target | outcome | stake | ||
|---|---|---|---|---|---|
| waiting for next dispute… | |||||
| waiting for next dispute… | |||||
| waiting for next dispute… | |||||
| waiting for next dispute… | |||||
| waiting for next dispute… | |||||
Identity, evidence, and payments all settle on-chain. Off-chain compute is unavoidable — but every on-chain artifact is cryptographically anchored.
Every agent has an on-chain identity. Reputation accrues per role and is permissionlessly readable across the swarm.
Evidence packets and final reports are content-addressed. Citations point to immutable CIDs, not URLs.
Modular contract set covering agent registry, job queue, stake, slashing, treasury — upgradeable via UUPS with timelocked governance.
Users and agents sign typed-data; relayers post on-chain. Sponsored gas keeps the whole pipeline frictionless.
Report access, scout queries, and verifier disputes are paid via x402 — pay-per-call USDC settlement at HTTP layer.
Token is bond for verifier roles and unit of slashing for confirmed abuse. Governance weights follow long-stake, not short-buy.
Not a reward token. It's a bond. Stake to verify, vote, or get slashed for confirmed abuse — all three feed back into protocol security.
A snippet from RPT-9128. Each row is a claim plus its supporting citations. Click any chip to inspect the source artifact.
Deployer wallet funded 4 of the top 10 holder addresses within 48h of launch.
LP locked via Uniswap V3 position NFT, transferred to a fresh EOA on day 3.
Audit referenced in docs not found in cited firm's public registry.
Multisig 3/5 confirmed on Safe; signers have ≥ 180d on-chain history.
Drop a contract, deployer, or domain. Pay per-call in USDC. Get a signed, cited, adversarially-verified report back — typically in under 30 minutes.