Three sources mention “Lumen Reserve.” Same company? Different companies? One of them sanctioned, the other not? The evidence is genuinely underdetermined — there is no string match, no rule, no LLM prompt that can resolve it from inside any single document.
Most pipelines force a binary answer at ingest. Kloom holds the evidence, scores the alternatives, and lets your workflow ask the identity question when it actually has the evidence to answer it.
A PKG isn’t a smarter database — it’s a different kind of object. One that reasons about identity, holds typed relationships through time, and stays live under the workflows querying it.
Probabilistic merging of mentions into entities — with confident identity, confident distinction, and genuine underdetermination held as three distinct states. No premature collapse.
Typed, directional, confidence-weighted relationships between entities. Attributes carry temporal bounds — what was true, when, and how sure we are.
PKG-QL standing queries running against the live graph. Subscribe to a mission; receive alerts the moment the evidence changes.
The PKG is the product — a four-layer graph that reasons about identity probabilistically. The work that makes it trustworthy happens beneath the surface: scoring matches, preserving evidence, and versioning every mutation so a query made today can be defended a year from now.
A four-layer graph — evidence, graph, cluster, curated backbone — that reasons about identity probabilistically. Resolve entities at query time under your own merge policy. Traverse typed relationships. Pin a session to a graph version so your workflow doesn’t shift under you mid-investigation.
tsconst entities = await pkg.query({
pkgql: `MATCH (e:Entity {name: 'Lumen'})
(company:Entity)
WHERE (e)-[:trades_with]->(company)
AND company.sanctioned = true
WITH PROBABILISTIC MERGE THRESHOLD 0.5
RETURN e, company`,
session_id,
});
// each entity carries:
// identity_distribution[]
// known_names[] (original script)
// corroborated_attributes[]
// evidence_sources[] Identity is stored as an evolving state, not a static fact. As new data is ingested, your view of it stays up to date.
Raw mentions are never discarded. Every entity claim points back to the document and passage that produced it.
Queries pin to a graph version for the lifetime of a workflow. Every mutation is recorded with the acting principal.
AML, KYC, and sanctions teams at banks, payment processors, and crypto compliance shops. Replace four rule-based tools with one evidence-backed platform.
Watch-list maintenance, sanctions screening, OSINT entity resolution, intelligence-context due diligence. The systems that ingest a million mentions a week and must defend every entity decision.
Identity work for national-security and FI customers can’t be outsourced to a vendor’s cloud. Kloom is a library you operate inside your perimeter, with the data flows you require.
Sovereign, integrator-owned, no phone-home. Runs on hardware-limited environments by design.
Every mutation — resolutions, anchor confirmations, predicate registrations — recorded with the acting principal.
Raw mentions are never discarded. Every entity claim points back to the document and passage that produced it.
Names are indexed in the script and language in which they arrive. No lossy index-time translation.
A demo runs against representative data from your domain and returns the resolved entities you can audit end to end. Thirty minutes, with engineering in the room.