Open-source sanctions screening platform. A free, open alternative to commercial watchlist screening solutions. Sieve fetches publicly available sanctions lists, normalizes them into a unified entity model, indexes them in memory, and exposes both a CLI and a REST API for screening names.

35 providers across 25 jurisdictions. All lists are fetched from official government endpoints, parsed into a unified entity model, and indexed in memory for screening.

graph LR
    CLI[sieve-cli] --> MATCH[sieve-match]
    VERTX[sieve-server] --> MATCH
    SPRING[sieve-spring-server] --> MATCH
    MATCH --> CORE[sieve-core]
    MATCH --> INGEST[sieve-ingest]
    VERTX --> INGEST
    SPRING --> INGEST
    CLI --> INGEST
    INGEST --> CORE

sieve/
├── sieve-core/              # Zero-dependency domain module
├── sieve-address/           # Address normalization (libpostal)
├── sieve-ingest/            # List fetchers and parsers
├── sieve-match/             # Matching engine implementations
├── sieve-server/            # High-performance Vert.x REST API
├── sieve-spring-server/     # Spring Boot REST API (PostgreSQL, Swagger, scheduling)
├── sieve-cli/               # Command-line interface
├── sieve-benchmark/         # Performance benchmarks
└── pom.xml                  # Parent POM

• Java 21+
• Maven 3.9+

mvn clean verify

# Fetch all enabled sanctions lists
java -jar sieve-cli/target/sieve-cli-0.1.0-SNAPSHOT.jar fetch

# Screen a name
java -jar sieve-cli/target/sieve-cli-0.1.0-SNAPSHOT.jar screen "John Doe"

# Screen with options
java -jar sieve-cli/target/sieve-cli-0.1.0-SNAPSHOT.jar screen "John Doe" --threshold=0.85 --list=ofac-sdn

# View index statistics
java -jar sieve-cli/target/sieve-cli-0.1.0-SNAPSHOT.jar stats

Exit codes: 0 = no match, 1 = match found, 2 = error (CI/CD friendly).

Two server implementations are available with identical API endpoints:

# Option 1: Vert.x (high-performance)
java -jar sieve-server/target/sieve-server-0.1.0-SNAPSHOT.jar

# Option 2: Spring Boot (full-featured)
java -jar sieve-spring-server/target/sieve-spring-server-0.1.0-SNAPSHOT.jar

The Vert.x server accepts CLI flags and environment variables:

java -jar sieve-server/target/sieve-server-0.1.0-SNAPSHOT.jar \
  --port 9090 --threshold 0.85 --eu true --uk true

# Screen a name
curl -X POST http://localhost:8080/api/v1/screen \
  -H "Content-Type: application/json" \
  -d '{"name": "John Doe", "threshold": 0.80}'

# List status
curl http://localhost:8080/api/v1/lists

# Refresh lists
curl -X POST http://localhost:8080/api/v1/lists/refresh

# Health check
curl http://localhost:8080/api/v1/health

# The EU list of prominent public functions: a summary, then one jurisdiction's functions of one category
curl http://localhost:8080/api/v1/pep/functions
curl "http://localhost:8080/api/v1/pep/functions/DE?category=a"

• Exact Match — Normalized case-insensitive exact comparison (score: 1.0 or 0.0)
• Fuzzy Match — Jaro-Winkler similarity (implemented from scratch, no external dependencies)
• Composite — Runs both engines, deduplicates by entity, keeps highest score

Sieve can look up recent news articles that name a person or organisation in a crime, corruption, sanctions or terrorism context, using the open GDELT news index. The articles are candidates for an analyst to read, never a match: they are kept apart from the curated lists, never stored as entities, never change a screening score or the screen exit code, and an empty result does not clear a name.

Two ways to ask GDELT:

• gkg (default): reads GDELT's Global Knowledge Graph files, which GDELT publishes every 15 minutes with the people, organisations and themes it found in the online news it processed, in English and machine-translated from other languages. Sieve keeps the articles tagged with adverse themes (money laundering, corruption, bribery or fraud, sanctions, terrorism, arrests, organised crime, trafficking and the like) for a rolling window and compares the searched name with the names in them, on the same matching key as list screening. Each file is about 5 MB in English and 12 MB translated, so the server reads a few hours at startup and then each new file as it appears.
• doc-api: asks GDELT's full-text search API per name, which covers the last three months. GDELT allows one request every five seconds and often refuses requests from shared cloud addresses, so Sieve answers UNAVAILABLE rather than queueing callers.

# CLI: read the last 6 hours of news files and look up names (exit 0 whatever it finds, 2 if GDELT could not be read)
java -jar sieve-cli/target/sieve-cli-0.1.0-SNAPSHOT.jar media "John Doe" "Acme Holdings"
java -jar sieve-cli/target/sieve-cli-0.1.0-SNAPSHOT.jar media --index doc-api --days 30 "John Doe"

# Spring Boot server, with sieve.adverse-media.enabled=true (off by default)
curl -X POST http://localhost:8080/api/v1/adverse-media \
  -H "Content-Type: application/json" \
  -d '{"name": "John Doe", "lookbackDays": 30, "maxArticles": 10}'

Each article carries its URL, headline, site, language, the date GDELT saw it, the adverse terms or themes that made it a candidate and, for gkg, the name in the article that was taken for the searched one. Names in GDELT are machine-extracted and an adverse article may be adverse for someone else it names, so every result needs reading.

• Vert.x server — configured via CLI flags and environment variables (see table above)
• Spring Boot server — see sieve-spring-server/src/main/resources/application.yml

A public dashboard at abgarsim.github.io/sieve-aml is rebuilt every night from all 36 lists: entity totals per list, a world map of sanctioned entities by nationality and address, data quality per source, benchmarks, and a searchable list of every record with a full data card. Politically exposed persons are counted on the dashboard but their records are not published there. The data comes from sieve snapshot; the site lives in dashboard/.

java -jar sieve-cli/target/sieve-cli-0.1.0-SNAPSHOT.jar snapshot --out snapshot   # write the data files
cd dashboard && npm install && npm run dev                                     # serve the site on that snapshot

# Spring Boot server (with PostgreSQL)
docker compose up sieve-spring

# Vert.x server (standalone, in-memory)
docker compose up sieve-server

• Java 21 — Records, sealed interfaces, pattern matching, virtual threads
• Vert.x 4.5 + Netty — High-performance server (event-loop, zero-copy I/O)
• Spring Boot 3.3 — Full-featured server (PostgreSQL, Swagger, scheduling)
• Picocli — CLI framework (no Spring dependency)
• StAX — Streaming XML parsing for large sanctions lists
• Jackson — JSON parsing
• Apache POI — XLSX spreadsheet parsing
• Playwright — Headless browser for JS-rendered sites (TR MASAK, IL NBCTF)
• JUnit 5 + AssertJ — Testing (parallel execution)

MIT — see LICENSE for details.