Skip to main content

TrustRelay Atlas

TrustRelay Atlas is a multi-tenant, graph-native KYB/AML investigation platform. It ingests data about companies and the people behind them from multiple providers, resolves that data into a single canonical knowledge graph, scores risk against configurable matrices, and produces auditable investigation reports.

This documentation describes the current system as it exists in the trustrelay-atlas repository. It is written for two audiences:

  • Engineers — who need to understand how the backend, ontology, pipelines, plugins, workflows, and frontend fit together.
  • Operators — who deploy, configure, monitor, and onboard tenants onto the platform.

The shape of the system in one diagram​

What makes Atlas distinctive​

  • Ontology-first. Everything resolves into a versioned ontology (currently v3.5) of LegalEntity, Person, Address, and their relationships. Providers never write directly to a fixed schema — they emit claims that are mapped, scored, and merged.
  • Claims and survivorship. Each attribute can have many competing claims from different providers. Survivorship rules — weighted by provider trust and guarded by protected fields — decide which claim wins, and every change is recorded as a mutation for audit.
  • Durable workflows. Long-running investigations are orchestrated as Temporal workflows, so a multi-minute, multi-module OSINT investigation survives restarts and is fully observable.
  • Strict multi-tenancy. Tenant isolation is enforced at the database layer with PostgreSQL Row-Level Security (RLS), not just in application code.
  • Configurable risk. Risk is computed both by per-module rules and by a configurable risk-matrix engine that operators can author without code changes.

How to read these docs​

If you want to…Start here
Understand what Atlas does end to endThe investigation story
See the full architectureArchitecture → System overview
Understand the data modelArchitecture → Data model
Call or extend the APIAPI → Overview
Deploy or run itOperations → Deployment
Look up a termGlossary
Accuracy & provenance

Pages cite concrete files and modules (for example src/api/main.py, schemas/ontology/ontology_v3.5.yaml). Code and SQL snippets are representative — they illustrate structure and intent rather than reproducing every column or line verbatim. When in doubt, the cited source file is the source of truth.