Investigation throughput · on autopilot · Est. 2026

Your investigators only get to a subset of the incidents that matter. We'd like to help with the rest.

A multi-agent platform on Merkle Science's transactions warehouse that quietly pre-investigates the cases your team doesn't have time for — so your analysts spend their hours on the parts only they can do.

Agent: Forensic Investigator · Case #MS-26-1149
LIVE
analyst@ Trace funds from 0xfd…8a3c over the last 90 days. Flag sanctioned counterparties and prep a SAR-ready summary.
Routing across 4 tools · sequence planned by the Forensic Investigator
01 tracker.trace_flow()depth=6 · 90d 4 hops · 84ms
02 compass.screen()OFAC · EU · UK · UN 1 hit · 28ms
03 kybb.entity_lookup()12 counterparties 12/12 · 41ms
04 dp.cluster_query()terminus wallets 3 clusters · 19ms
High Risk
Findings. 38.2 ETH (≈ $142K) routed via Tornado Cash; terminus cluster attributed to Garantex (OFAC SDN, 2022). Confidence 94%. SAR draft ready for review.
§ 00 / The maths
The cost
~6h
Senior investigator time per case on preliminary work alone — pulling addresses, sketching the fund flow, profiling contracts.
The reality
Ship the top 10%
By gut-feel severity. The remaining 90% is a backlog you never reach.
The fix
Multi-agent
A platform on Merkle Science's transactions warehouse that does the preliminary work for you.
Operating across Tier-1 Exchanges Custodians Government Agencies Financial Institutions Law Enforcement
§ 01 / The shift

Stop running the playbook. Start reviewing the work.

The investigations workflow hasn't changed in a decade. The way it gets done is about to.
Today
  • Trace funds across 6 hops.
  • Diligence a token.
  • Pull a cohort.
  • Draft a SAR.
  • Build a graph.
With Studio
Data Analyst Forensic Investigator Token Listing Agent
Three agents. Shared memory. One audit log.
§ 02 / The agents

Three specialists. One swarm.

Each agent handles a workflow on its own — or they compose into a multi-agent investigation, planned and routed for you.
Planning layer

A planner routes a case through whichever specialists make sense, then synthesises a draft your analyst reviews — not writes from scratch.

01

The Data Analyst

Pulls the cohort. Runs the cut. Charts the answer.

Translates a question in English into a Data Platform query, runs it across the attribution warehouse, and derives the supporting metrics — concentration, velocity, period-over-period change. Returns charts, dashboards, or inline tables, with the SQL attached so you can sign off on the methodology.

hours → minutes
analyst@ Cohort all wallets that touched Tornado after the SDN.
dp.cluster_query() cohort + chart SQL
02

The Forensic Investigator

Traces every hop. Drafts the filing.

Follows funds in either direction — through mixers, across bridges, swaps, and into terminus clusters — and turns the trace into a SAR-ready memo with citations inline. Exports the case to CSV when the regulator wants the raw ledger.

days → minutes
analyst@ Trace 0xfd…8a3c outbound, depth 6, last 90 days, and draft the SAR.
tracker.trace_flow() compass.screen() report.draft()
03

The Token Listing Agent

Diligences a token before you list it.

Reviews contract history, holder distribution, liquidity, sanctions exposure, issuer background, and audit history. Returns a listing-grade memo with the red flags, green flags, and the evidence behind each one — plus charts the listing committee can read at a glance.

days → minutes
analyst@ Diligence $XYZ before Friday's listing decision.
token.profile() kybb.contract() liquidity + charts
§ 03 / In · Out

How work moves through the system.

Four ways cases enter. Three ways they exit. The swarm handles the bit in the middle.
→ Four ways cases enter
  1. 01
    Autonomous OSINT discovery. Exploit briefs ready before the headline trends.
  2. 02
    Scheduled monitors. Recurring jobs on watched wallets and counterparties.
  3. 03
    REST API. Pipe your existing case queue or CMS straight in.
  4. 04
    In-chat prompt. Analyst types a question; the swarm answers it.
→ Three ways they exit
  1. 01
    In-app brief. Plus fork-to-workspace for hands-on continuation.
  2. 02
    Email. Alerts on material events · daily / weekly org digest.
  3. 03
    Webhook to your stack (v1.5). CMS, Slack, Linear — pushed where you already work.
§ 04 / How it works

Three steps. No ceremony.

Studio is a thin conversational layer over infrastructure that already exists. Investigators bring the question; the agents bring the data and the reasoning.
STEPi

Ask

Drop in addresses, hashes, alerts, case files. The agent picks the right tools.

STEPii

Work

The agent plans, calls Compass, Tracker, KYBB, the Data Platform. Pulls in another specialist when the case demands it.

STEPiii

Review

Cited answer, graph view, draft filing. You sign or send back for another pass.

§ 05 / Data fabric

Agents are only as sharp as the data underneath.

Real-time chain coverage, resolved bridges, attribution clusters, and connected sanctions feeds — across the jurisdictions our customers operate in.
Chains indexed
30+
Indexed in real time, with new chains shipping faster than the industry average.
Assets covered
10K+
Tokens, NFTs, and digital assets covered with full transaction history.
Bridges resolved
200+
Resolved automatically — traces never stop at the chain boundary.
Attribution points
1B+
Entities, clusters, behavioral signals, and sanctions hits.
Connected sources
Compass API Tracker API KYBB API Data Platform OFAC SDN EU Consolidated UK OFSI UN Sanctions Attribution Clusters Threat Intelligence DeFi Protocol Decoder VASP Risk Scores Darknet Monitoring Custom Webhooks
§ 06 / Target outcomes · 12-month horizon

Where pilot orgs land in year one.

Honest targets, not marketing inflations. Measured per pilot organisation across the first twelve months.
Metric Before Target with AI Studio
Incidents investigated Top 10% 100% pre-briefed
Avg. senior-analyst time per case ~6 hours ~1 hour, review & finalize
Token-listing review turnaround 3 days Same day
Cases per FTE per quarter ~30 ~150
Differentiation

Built on Merkle Science's transactions warehouse and tag taxonomy. Every agent uses the same data your analyst already trusts — not a wrapper on a public LLM.