Cypher-nativeQuery language
Define, mutate, and query connected data directly in its native model.
Latest release v0.2.0→
The one data index for your agentic applications
Graph, vector, and full-text retrieval over the same transactional data—embedded in your application or running behind a service.
PythonNode.jsCLI
Data foundation
A transactional data foundation
NeuG manages data before it indexes it. A graph-native engine owns schema, transactions, persistence, and recovery—so every retrieval path starts from one consistent state.
ACIDTransactions
Keep reads and writes consistent as application state changes.
Two modesEmbedded · Service
Run in-process or expose the same data engine behind a service.
ExtensibleExtension system
Add new capabilities through extensions while keeping the core focused.
Apache 2.0Open source
Use, modify, and distribute NeuG under the Apache 2.0 license.
Unified retrieval
Three indexes. One result
Query structure, semantics, and exact keywords together. NeuG ranks one result set without moving data across engines.
QueryPaymentService+"retry timeout"
Ranked result · #1RetryPolicy
Ranked result · #2TimeoutError
Ranked result · #3Runbook
Graph algorithms
Analyze the graph in place
Retrieval finds relevant entities. Graph algorithms reveal importance, communities, and patterns across the wider structure.
The same graph data as the retrieval example above
Reveal influence
See which entities matter most across the graph.
Beyond PageRank and LeidenRun a broader set of graph algorithms over the same data.Explore graph algorithms →
Why NeuG
Choose the right architecture
Compare NeuG in the deployment mode that fits your workload—or against the multi-engine stack it can simplify.
Service modeNeuG vs Neo4j
Throughput · QPS ↑Embedded modeNeuG vs LadybugDB
Query latency ↓Service · 50.6× throughput
617 QPS for NeuG vs 12.2 QPS for Neo4j across 14 LDBC SNB Interactive complex reads.
Embedded · faster on 8 of 9
Single-threaded NeuG beat LadybugDB's best multi-threaded result on eight of nine LSQB queries.
One engine, two deployment modes
Run in-process for low-overhead analytics, or expose the same runtime as a service for concurrent workloads.
—
REPRODUCIBLE DUAL-MODE BENCHMARK
LDBC SNB SF1 · Apple Silicon · read workloads. Service test: 4 clients for 300 seconds. View methodology →
Quick start
From install to first query
NeuG runs where your application runs. Install the library, open a database, and query it with Cypher.
$ pip install neug
import neug
db = neug.Database("")
conn = db.connect()
result = conn.execute("""
MATCH (n) RETURN n LIMIT 10
""")Compatibility
- Data formats
- Apache Arrow · Parquet
- Object storage
- S3 · OSS
- Operating systems
- Linux · macOS
Windows · Roadmap - Architectures
- x86_64 · ARM64
One index. Fewer moving parts
Start embedded. Move behind a service when your application needs it.