Architecture note · Product + research

Labeled Meta Property Graph: what RushDB implements, and what remains research.

RushDB keeps application values on records while representing field definitions as queryable graph metadata. This page separates that shipping architecture from broader claims about future indexing, execution, and performance.

Product meaning: schema is not only documentation outside the database. The fields and types observed on records contribute to a graph-backed structural model that software can inspect at runtime.

Start building See the system architecture

Model

Property metadata is part of the graph.

RushDB implements LMPG on top of Neo4j. Neo4j provides storage and transaction mechanics. RushDB’s model keeps each record’s scalar values on that record and connects it to metadata describing the property definitions it carries. Property identity includes both name and inferred type.

Record

The application entity. Scalar values remain on the record as the source of truth.

Property definition

A graph metadata node identified by property name and inferred type, connected to records that carry it.

Relationship

An explicit directed connection between records, independent from the property metadata layer.

Queryable schema

Labels, field definitions, value domains, and relationship paths exposed through schema and property APIs.

Worked example

Observe the model through its public schema surface.

Internal metadata nodes are an implementation detail. Applications and agents use supported schema, property, value, and query APIs rather than issuing queries against RushDB’s internal labels.

Inspect the model

Read the schema software will use to form queries.

The schema surface turns observed labels, typed properties, and relationship paths into machine-readable context. It is a projection of stored structure, not a separate hand-maintained schema file.

Request

schema = db.ai.get_schema_markdown({
    'labels': ['ACCOUNT', 'FEATURE', 'EVENT', 'INCIDENT'],
})

print(schema.data)

What to notice

RushDB’s LMPG layer exposes property usage and relationship structure over Neo4j-backed storage.
Read the returned Markdown rather than assuming labels or paths; schema freshness and cache behavior are deployment concerns.

Implementation matrix

Separate availability from architectural ambition.

Status	Scope	Public claim boundary
Implemented	Flexible labeled records, typed property metadata, nested relationship creation, Schema API, property/value discovery, SearchQuery, transactions.	Safe to describe as current product behavior.
Configuration-dependent	Managed or external embedding indexes; LLM-assisted Smart Search; suggested relationship analysis.	Requires an index, model/provider configuration, or explicit approval depending on the capability.
Neo4j-constrained	Storage layout, transactional mechanics, index support, and low-level graph execution.	RushDB adds the product model and API but does not replace the underlying engine’s physical constraints.
Research direction	Edge-localized vector indexing, generalized performance advantages, storage-efficiency claims, and alternative engine-level execution strategies.	Do not treat as a generally available feature or benchmark result.

Open questions

What this page does not claim.

A property-centric model is useful without turning unmeasured hypotheses into promises. These questions require controlled benchmarks, implementation evidence, or additional engine work.

Storage overhead of the property metadata layer has not been published as a controlled benchmark.
No public empirical study currently proves data-size-independent schema discovery latency or universal I/O reduction.
Vector-on-edge and edge-scoped ANN concepts remain distinct from the current managed/external property embedding indexes.
Transactional behavior should be evaluated through the current RushDB and Neo4j implementation, not inferred from a conceptual graph diagram.
Suggested relationships are proposals. Shared fields do not silently become domain edges without an explicit or approved mutation.

Primary sources

Read the public implementation references.

This is a product architecture and research-boundary note, not a peer-reviewed benchmark paper. The public documentation is the authority for shipping behavior; earlier essays are historical context and may contain broader hypotheses.