Investigation Desk

The powers of an elite on-chain sleuth — for anyone who can ask a question.

Sleuth AI runs the whole forensic workflow on-chain in plain language — the moves a senior analyst would make, in one conversation, with evidence and a confidence label behind every claim.

The Problem

On-chain, information asymmetry is being weaponized.

The truth is public — every transaction, on an open ledger. But only a handful of experts can actually read it. That gap is the grift: bad actors hide behind the complexity, and everyone else pays for it.

Scammers & rugs

Deployers shill a token, dump on buyers, and vanish — then redeploy under a fresh wallet. Hidden airdrop rules, insider exits.

Bad-faith founders

Fake community and "fair" launches with airdrop bait to draw a crowd — engineered for a quick rug, not a real product.

Crooked influencers

A bag casually shilled to followers, then quietly dumped into the buy pressure it just manufactured.

Snipers & bundlers

Bots and bundled buys faked as organic demand — the loud cases are only the ones that got noticed.

State-level actors

Stolen funds laundered through chains, bridges and mixers — betting no one has the patience to follow.

A handful of elite on-chain sleuths can see through all of it. Everyone else is flying blind.

The Product

A sixth sense for on-chain — by just asking.

Sleuth AI runs investigation and due diligence on-chain in plain language — the whole forensic workflow an elite sleuth would run, in one conversation. It closes the asymmetry: everyone gets to read the truth.

01 · Ask

Plain language, or @account $ticker /skill

02 · Investigate

The agent plans and runs the forensic workflow end to end.

03 · Dossier

Findings gathered into one evidence-backed case file.

04 · Share

Publish the trail — others can read and contribute.

patternevidenceconfidencecaveat / next pivot
Ask It Anything

Investigation, in plain English.

The workflows analysts run by hand across a dozen tools — Sleuth runs them for you, then chains them together.

>give me the complete funding path of [wallet]funding chain
>what does this transaction do? [tx]contract reading
>detect pump-&-dump activity around [social post]manipulation
>find wallet candidates for [social account]attribution
>who is the entity behind [ticker]?entity recognition
>is this contract safe to interact with? [address]safety
>surface every side wallet linked to [wallet]side wallets
>who were the first 10 buyers of [ticker]?insider overlap
>what influencers spoke about [ticker]?social signal
Under the Hood

Powered by an agentic deep dive into on-chain data.

Not a dashboard with an AI wrapper. A planning agent that decides what to query next, recursively, until the trail resolves.

Plan

Resolves intent into a forensic plan — the same moves a senior analyst would make.

Deep-dive

Runs queries, reads results, and pivots — hop after hop — across raw on-chain data.

Ground

Every step backed by evidence with a confidence label. No black-box assertions.

$TICKER · 7D · events anchored the answer picks its own shape
wallet buysocial postmarket dump
wallet move social market
Drop-anything ingestion

Paste a screenshot or a social post link — vision reads it, extracts the $tickers and 0x… addresses inside, and opens the investigation.

Entity recognition

Resolves a wallet to its on-chain identity — ENS & Basenames, forward and reverse, plus linked public social accounts — so an address becomes a recognizable entity.

Parallel fan-out

One question fans into many sub-investigations that run in parallel, then merge — early-buyer portfolios, side-wallet trees, holder overlaps — at machine speed.

Trust & Accuracy

Every claim carries a receipt.

Forensics is only useful if it's defensible. Sleuth is built to be checked — not believed.

Evidence-linked

Every finding traces back to on-chain facts you can open and verify yourself. No black-box assertions.

Confidence-labeled

Each step is graded by confidence. Probabilistic links are flagged as such — never dressed up as certainty.

Observations, not verdicts

The agent reports what the data shows and lets you conclude — it won't fabricate a label the evidence can't support.

Eval-gated

Capabilities are graded against a growing eval suite before they ship — accuracy is measured, not assumed.