An agent that investigates the wallet instead of just assigning a score
It picks its own tools, follows the money across hops, cross-checks five data sources and collects evidence. The result is a risk score where every factor has a transaction, a source and a timestamp.
Demo scenario.
A think, check, record loop
At each step the agent forms a hypothesis, picks a tool, reads the response and decides what to do next. If the trail is clean (an exchange, a hosted wallet, a mining pool), it stops at step 3 or 4. If it finds risk, it digs deeper.
Everything is visible in real time: the agent's reasoning, checks and new evidence appear in the interface as it works.
Hypothesis
“Incoming funds come from addresses that look like a mixer. Exposure needs checking.”
Check
Requests the wallet's exposure and cross-checks the answer with a second source.
Continue or finish
Follows the money to the next wallet or wraps up the investigation.
Evidence
Saves the finding with its type, severity and hash: [E1].
Leading blockchain data providers are already connected
The agent works with one provider or all of them at once, cross-checking their answers. Plus our own attribution database and clustering.
Chainalysis
connectedGlobal Ledger
connectedBreadcrumbs
connectedArkham
connectedIn-house intelligence
built inA team of agents instead of one
In multi-agent mode, an orchestrator hands out work to specialists. The first wave runs in parallel. The counterparty tracer joins only if risk is found. The second wave assembles evidence and prepares an STR recommendation.
The address receives funds from two clusters with the same exchange name. Treat them as one organization?
Asks when data is missing
In interactive mode, the agent can pause and ask the analyst a question: to clarify an entity, the scope of the check, a classification or a priority. Every question has a default answer, so the investigation never stalls.
The agent gathers facts. A formula calculates the score
The final score isn't made up by a language model. A deterministic engine calculates it: the same data always produces the same result, and every score can be broken down into its components.
Spots bots and structuring
Suspiciously even intervals between transfers, amounts just under the $10k, $5k, $3k and $1k thresholds, the share of round amounts, plus a heatmap of activity by hour of day.
Finds every address the owner controls
Real-time common-input heuristic, with safeguards against CoinJoin and exchange batch payouts.
Keeps watching and alerts you when it matters
Wallet monitoring: rules turn events into signals, and the AI decides whether to raise an alert. Alerts arrive in the interface and by email.
Every sentence links to evidence
The agent's conclusion is tagged with references like [E1] and [E2]. Each one leads to a transaction, a source and a collection time. The report exports to PDF with a link graph, fund flows and a transaction appendix, in Russian and English.
From investigation to STR →The wallet received funds directly from a mixer [E1] and sent them to a counterparty on the OFAC SDN list [E2]. Amounts stay close to the $10,000 threshold [E3].
- [E2]
- Counterparty on a sanctions list
- Source
- OFAC SDN sanctions list
- Transaction
- [transaction hash]
- Severity
- CRITICAL
See the agent on your wallets
Send us a few addresses. We'll go through them together and show every step of the investigation.