Chapter 22Future Enhancements

This chapter sketches plausible directions for the evolution of GQL beyond the current standardization snapshot, focusing on features that improve compositionality, interoperability, and practical deployability. We highlight enhancement themes that recur across vendor roadmaps and the research literature—graph-to-graph query composition, richer schema and constraint support, and extensions of the underlying data model. Finally, we outline how these directions connect to emerging needs around streaming graphs and AI-enabled workloads, while preserving the declarative and analyzable character of the language.

22.1 Compositional Enhancements

The ongoing development of GQL presents numerous opportunities to enhance its capabilities, particularly by introducing compositional features such as graph-returning subqueries and views defined using new operators for graph projection and combination (Deutsch et al. 2022; Francis et al. 2023a; Gheerbrant et al. 2025b; Kondylakis et al. 2025). These enhancements enable more complex, efficient, and secure data interactions. This section explores potential advancements in these areas and their implications for GQL.

GQL already supports nested query specifications, including correlated uses that can refer to bindings from an enclosing query. The principal compositional gap is therefore not ordinary table-producing subqueries but graph-returning subqueries whose results can serve as graph inputs to later query parts. Potential advancements include recursive graph-returning subqueries, which could support reusable fixpoint-style graph computations, and explicit subquery materialization for results that are reused within a larger query (Angles et al. 2017; Deutsch et al. 2022).

Views in GQL can provide a powerful mechanism for abstracting and reusing query logic, enhancing developer productivity and query performance. From a systems perspective, implementing property-graph views forces explicit choices between virtual evaluation and materialized storage, and it raises the need for view-aware query rewriting and indexing strategies. Recent work has developed and experimentally compared such implementation strategies across both native property-graph engines and relational backends, clarifying the performance trade-offs that arise when views become first-class and reusable program units (Han and Ives 2024; Han and Ives 2025). Moreover, when views are materialized (or partially materialized), incremental view maintenance becomes central; certified maintenance frameworks based on Regular Datalog demonstrate how to derive delta-style maintenance programs with formal soundness guarantees, which is directly relevant when view results must remain trustworthy under frequent graph updates (Bonifati et al. 2018). Potential enhancements in this area include parameterized views, which enable views to accept input parameters, allowing for dynamic query construction and reusability across different contexts. Updatable views could permit modifications to the underlying graph data through views, simplifying data manipulation tasks. View composition would enable the creation of complex views by combining existing views, promoting modularity and code reuse (Angles et al. 2017; Deutsch et al. 2022).

Recent theoretical analyses of GQL and related graph querying facilities have also highlighted that current property graph query languages are, in practice, mostly graph-to-relation formalisms and therefore fall short of full compositionality at the graph level (Francis et al. 2023a; Bonifati 2025; Gheerbrant et al. 2025b). Future extensions of GQL could address this gap by providing a closed algebra of graph-to-graph operators whose results are again graphs and can be fed as inputs to subsequent queries (Angles et al. 2017; Gheerbrant et al. 2025a). Such an algebra would enable principled reasoning about query equivalence, rewriting, and optimization in a way that mirrors the role of relational algebra for SQL, while still respecting the semantics of path expressions, graph constraints, and schema conformance.

Recent work sharpens this limitation by showing that GQL and SQL/PGQ, despite being able to express reachability and all first-order queries, do not capture all queries in $\mathrm{NLOGSPACE}$. The proposed language couples regular path queries with variables and data-value comparisons to $\#\mathrm{Datalog}$, a fully compositional graph-to-graph language that can construct nodes, edges, lists, and paths, and provides a concrete extension of both standards (Arenas et al. 2026).

A future graph-builder construct could offer graph-valued construction and transformation capabilities beyond the current standard. Such an extension would enable users to perform more sophisticated operations on graph data, facilitating tasks such as data integration, data warehousing, and graph-based machine learning applications. Potential features include conditional node and edge creation, which allows the creation of nodes and edges based on specific conditions, enabling dynamic graph construction. Batch operations could provide mechanisms for creating multiple nodes and edges in a single operation, improving efficiency for large-scale graph construction. Graph transformations would introduce functions for modifying existing graph structures, such as merging nodes, splitting edges, or applying graph algorithms (Angles et al. 2023; Kondylakis et al. 2025).

A concrete design point for such a facility uses GPC-based rules to construct output property graphs while accounting for property values rather than topology alone; for transformations using only simple and trail path restrictors, consistency checking is $\mathrm{PSPACE}$-complete. A proof-of-concept implementation in openCypher shows that declarative transformation rules can be translated into executable scripts with practical advantages over ad hoc transformation scripts, suggesting that a GQL graph builder should combine explicit constructors with run-time conflict detection and implementation-aware optimization (Bonifati et al. 2024).

22.2 Schemas in GQL

The ongoing development of GQL aims to bridge the gap between the current graph types and the complex schema requirements necessary for practical data graphs. This section outlines potential enhancements to the schema capabilities of GQL, drawing on the experiences and expertise of users, vendors, and researchers (Angles et al. 2021; Angles et al. 2023; Kondylakis et al. 2025).

The current GQL standard includes optional graph types that encompass sets of node types and edge types, together with key and unique constraints. However, the experience of enterprise graph deployments indicates that additional schema features are necessary to describe and prescribe the structure and values of practical data graphs, akin to the complexity found in SQL schema (Angles et al. 2023). Essential conceptual modeling and schema definition features include domains for user-defined refinements of data types (such as ranges and patterns), the ability to specify nested record types for graph element attribution, and the capacity to define subtyping relations for attribution types. Additionally, features like default and computed values, richer key modes, referential, cardinality, and participation constraints for entities and relationships, and support for $N$-ary relationships are desirable (Angles et al. 2021; Angles et al. 2023).

Community proposals for property graph schema systems already demonstrate that such rich schema facilities are both feasible and practically useful in real-world deployments (Angles et al. 2021; Angles et al. 2023; Manouvrier and Belhajjame 2024). They combine flexible typing with multi-inheritance, a spectrum of key modes for nodes, edges, and properties, and a unifying treatment of integrity constraints. Future revisions of GQL could explicitly align the schema language with these developments, for instance by standardizing key modes, reusable constraint templates, and mechanisms to relate graph types and schemas in a single DDL framework.

Schema evolution should also support controlled movement along the spectrum between descriptive and prescriptive schemas. One proposed framework models conformance by a homomorphism from an instance graph to a schema graph and models evolution by graph-rewriting operations propagated between the two levels, suggesting high-level GQL schema-manipulation operations for both data-driven schema adaptation and schema-driven data migration (Bonifati et al. 2019).

These prospective constraint templates can be grounded in the theory of graph dependencies, which provides reusable constraint schemata ranging from key- and equality-style constraints to generating dependencies that can prescribe the existence of additional graph structure when certain patterns are present (Fan and Lu 2019; Shimomura et al. 2020).

There are also graph-specific schema requirements that need to be addressed. These include the ability to specify sub-graphs as integral units for insertion or maintenance of data graphs, for access control purposes, or as the target of constraints. Another critical feature is schema inclusion or “schema unions”, allowing the definition of nodes in a schema graph that can be the same or forcibly different from nodes of the same type in another schema graph. This capability facilitates the composition of schema graphs into larger, modular schemas and supports the creation of “simple sub-graph views” or “composite graphs” (Angles et al. 2023; Kondylakis et al. 2025).

The goal of extending the GQL schema language is to support a more restrictive set of constraints on the permissible values of GQL property graphs than currently allowed by graph types. The proposed schema definition language should be expressed as GQL DDL statements that either supplement or replicate the meaning of the existing DDL statements for graph-type definition and maintenance. This language will allow the creation, alteration, and deletion of graph schemas in a GQL catalog, including managing the relationship of an instance (data) graph to a graph schema throughout its lifecycle (Angles et al. 2021; Angles et al. 2023).

A comprehensive schema language for GQL should address the definition of schema conformance and the efficient maintenance of conformance during transactional updates. For instance, DML statements can explicitly reference schema sub-graphs to reduce the validation scope when executing mutations. The schema language should also support partial (extensible) schemas, allowing a “sliding scale of schema” that ranges from minimal constraints to schemas as or more restrictive than SQL schemas (Angles et al. 2023; Kondylakis et al. 2025). An extended GQL schema, possibly in conjunction with extensions to the data model (see below), has the opportunity to close the gap between graph schemas and attributed entity-relationship models, which permit $N$-ary relationships. This would more precisely align existing record-oriented conceptual data modeling techniques, such as Extended Entity Relationship Modeling or UML class diagrams, with graph schemas. These mechanisms could further interact with automated schema discovery and exploration tools for property graphs (Lbath et al. 2021; Bonifati et al. 2022).

22.3 Extending the Data Model and Integrating AI

The future of GQL holds great promise for advancements in data models and AI, enabling the development of sophisticated applications that leverage the full potential of graph databases. This section outlines potential directions for future versions of GQL, focusing on extending data models and integrating AI to enhance the language’s capabilities and applicability (Angles et al. 2017; Kondylakis et al. 2025).

It is essential to explore extending the property graph data model in GQL to include features such as graph properties, stored paths, hyperedges, multiproperties, and support for first-class (multi)properties and property values (Kondylakis et al. 2025). Graph properties enable annotating and organizing graphs in the catalog with additional rich metadata. Stored paths enable the storage and querying of frequently accessed routes within the graph, which is particularly useful in applications like social network analysis and recommendation systems. Hyperedges allow connections between more than two graph elements, providing a more expressive data model for representing complex relationships. Multiproperties enable multiple properties with the same name on a single node or edge, offering greater flexibility in data modeling. Support for first-class (multi)properties ensures that every multiproperty as well as every individual property value can be treated as a node, to allow their occurrence as the endpoints of edges in the graph topology and further (recursive) annotation with additional meta-properties. Finally, there are other relevant directions toward further development of the property graph data model landscape in extensions for temporal, geospatial, and probabilistic graph data, as well as the integration of property graph data management and stream processing (Kondylakis et al. 2025).

In particular, integrating streaming capabilities directly into GQL raises the prospect of a unified model for querying both static and continuously evolving property graphs. Future versions of the language could standardize windowing constructs, continuous query semantics, and incremental maintenance of graph views, so that streaming and at-rest graph data can be queried and optimized within a single conceptual framework rather than through ad hoc extensions (Rost et al. 2024; Kondylakis et al. 2025).

Standardizing concrete privilege-management and role-based access control (RBAC) statements in GQL can also enhance security and make user-permission management portable. RBAC allows for fine-grained control over access to graph data, ensuring that only authorized users can view or modify sensitive information. This is crucial for maintaining data integrity and confidentiality, especially in collaborative or multi-tenant environments.

Future versions of GQL should consider proposing statements for dynamically altering graph types, ensuring flexibility in adapting to evolving data requirements. This feature would enable users to modify the graph schema without recreating the entire graph, facilitating seamless schema evolution and minimizing disruptions to existing applications. Dynamic graph-type alteration can support a broader range of applications by allowing more adaptable and responsive data management practices (Angles et al. 2021; Angles et al. 2023).

Integrating vector search, knowledge graphs, and data science within the GQL framework can pave the way for building breakthrough Generative AI (GenAI) applications. These applications can deliver highly accurate responses, rich context, and deep explainability by leveraging the power of graph representations and AI models. Graphs can effectively represent hidden patterns and relationships within data, enabling GenAI models to make sense of the world and understand a specific company’s data and view of the world.

GQL already provides optional vector types and vector operations, including VECTOR_DISTANCE and VECTOR_NORM. The remaining opportunity is to integrate these values and operations with portable vector indexes, nearest-neighbor search, ranking, and query optimization. Such integration can improve the accuracy of search results and enable more effective information discovery, particularly in applications involving large and complex datasets, while complementing traditional graph querying with semantic relevance ranking.

A further opportunity lies in making AI techniques, such as graph embeddings and large language models, schema-aware. Current work on embeddings and generative models for graphs suggests that incorporating structural constraints and schema information remains challenging, yet essential for accuracy and interpretability (Hamilton et al. 2017). Tight integration of GQL with a rich schema language and constraint system could enable embeddings and GenAI components that respect keys, cardinalities, and type hierarchies by construction, and that can be explained back in terms of the underlying graph schema.

Incorporating data science techniques into GQL can enable advanced analytics and predictive modeling on graph data. By leveraging machine learning algorithms and statistical methods, GQL can uncover deep insights from the data, identify trends and patterns, and provide more informed decision-making. This integration can also enhance the capability of GenAI applications by combining structured graph data with powerful analytical tools (Hamilton et al. 2017).

The synergy of vector search, knowledge graphs, and data science can lead to the development of breakthrough GenAI applications that offer highly accurate responses, rich context, and deep explainability. These applications can leverage the interconnected nature of graph data to provide more meaningful and relevant insights, enhance user experiences, and drive innovation in various domains. For example, GenAI can utilize graph data in personalized recommendation systems to comprehend user preferences and relationships, resulting in more precise and tailored suggestions.

By incorporating these advancements, future versions of GQL can empower developers and data scientists to build more sophisticated, secure, and flexible applications that leverage the power of graph data models and AI. This will enhance the functionality and usability of GQL, enabling the development of next-generation intelligent systems that provide deeper insights and more effective solutions across various domains.