ISONGraph is a token-efficient property graph database designed for AI and LLM applications. It stores nodes, edges, and properties in the ISON format, achieving 70% token reduction compared to JSON-based graphs while maintaining 92% accuracy on graph reasoning tasks.

ISONQL is the query language for ISONGraph, similar to Cypher for Neo4j or Gremlin for Apache TinkerPop. It supports pattern matching, traversals, aggregations, and graph algorithms with token-efficient syntax optimized for LLM context windows.

Graphantic is the Pydantic-style schema validation system for ISONGraph. It provides type-safe node and edge definitions, automatic validation, and seamless integration with LLM output parsing for knowledge graph construction.

How does ISONGraph compare to Neo4j for LLM applications?

ISONGraph is specifically optimized for LLM context windows, using 70% fewer tokens than Neo4j's JSON export format. It's designed for in-context graph reasoning rather than persistent storage, making it ideal for RAG pipelines and multi-agent AI systems.

Can ISONGraph be used with ChatGPT, Claude, and Gemini?

Yes, ISONGraph is designed to work with all major LLMs. Its token-efficient format allows larger knowledge graphs to fit within context windows, and LLMs can directly read, reason about, and generate ISONGraph structures.

ISONGraph - Token-Efficient Property Graph Database for AI & LLMs

The Challenge

Graph Data Eats Context Windows

Traditional graph formats like JSON, GraphML, and RDF are verbose and waste precious context space. When your LLM needs to reason over relationships, every token counts. ISONGraph uses a compact tabular format that LLMs naturally understand from their training data.

68.6%

Token Reduction (Benchmarked)

90%

Knowledge Graph Accuracy

92%

Data Traversal Accuracy

vs 9 Other Formats

Complete Ecosystem

Everything You Need for Graph AI

From core graph operations to query language to schema validation.

ISONGraph Core

Property graph store with O(1) lookups

Features: CRUD, traversal, path finding Languages: Python, TS, Rust, C++ Use for: Knowledge graphs, RAG

ISONQL

Query language for graph operations

Features: NODES, EDGES, TRAVERSE, PATH Style: SQL-like + Fluent API Use for: Complex queries, filtering

Graphantic

Schema validation library

Features: Type validators, constraints Like: Pydantic + Zod for graphs Use for: Data integrity, LLM output

Format Comparison

Benchmark Results

ISONGraph was tested against 9 other formats across 150 questions in two benchmark suites.

Knowledge Graph Benchmark

100 Questions

Format	Tokens	Accuracy	Efficiency
ISONGraph	1,698	90.0%	53.00
ISON	1,976	88.0%	44.53
JSON Compact	2,893	88.0%	30.42
Cypher	3,522	89.0%	25.27
JSON	5,406	87.0%	16.09

                            Data Traversal Benchmark
                            50 Questions
                        
                                        Format
                                        Tokens
                                        Accuracy
                                        Eff (Acc/1K)
                                    
                                        ISONGraph
                                        639
                                        92.0%
                                        143.97
                                    
                                        ISON
                                        685
                                        88.0%
                                        128.47
                                    
                                        TOON
                                        856
                                        80.0%
                                        93.46
                                    
                                        JSON Compact
                                        1,072
                                        82.0%
                                        76.49
                                    
                                        JSON
                                        2,039
                                        84.0%
                                        41.20