Evokoa released pgGraph, an open-source PostgreSQL extension that adds graph database functionality to existing Postgres tables without requiring data migration. Released on GitHub on May 13, 2026, the extension gained 140 stars and 11 forks within three days, demonstrating strong developer interest in graph capabilities that work directly with standard relational data.
Built Specifically for AI Agent Workloads
pgGraph addresses a specific challenge faced by AI agents: traversing millions of relationships across databases without copying data to a separate graph system. The extension maintains PostgreSQL as the authoritative data source while building a derived graph index that enables relationship queries through SQL functions in the graph schema.
Evokoa designed the tool to sit between AI agents and live databases, using a core path built in Rust for speed and predictability. The system keeps relationships in memory, traverses them quickly, and fetches rows from the source database only when needed, supporting both SQL and JSON API access.
Technical Architecture Uses Memory-Mapped Structures
The extension employs Compressed Sparse Row (CSR) data structures for O(1) neighbor lookups and memory-mapped files for efficient caching. Built with pgrx, a Rust framework for PostgreSQL extensions, pgGraph operates as a graph-native memory layer compiled from relational metadata. This architecture enables fast traversals without recursive SQL queries, which can be slow and resource-intensive in traditional relational databases.
Key capabilities include:
- Graph search and traversal across related records
- Shortest path calculations between nodes
- Relationship queries without a separate database
- Support for PostgreSQL versions 13-18
- Explicit circuit breakers including depth limits, visited-node tracking, frontier limits, pagination, and strict memory safeguards
Early Stage with Production Use Discouraged
The project is currently at version 0.1.0 in early alpha stage. While tested for stability, Evokoa discourages production deployments at this time. The extension differentiates itself from existing alternatives like Apache AGE and AgensGraph by operating on existing tables without data migration and targeting AI agent use cases specifically.
Your standard PostgreSQL tables retain all normal features including constraints, indexes, backups, and row-level security. The graph functionality layers on top without requiring schema changes or data movement, making it particularly attractive for teams already invested in PostgreSQL infrastructure.
Key Takeaways
- pgGraph adds graph database capabilities to existing PostgreSQL tables without requiring data migration or a separate graph database
- The extension gained 140 GitHub stars and 11 forks within 3 days of its May 13, 2026 release
- Built in Rust using pgrx with Compressed Sparse Row data structures for O(1) neighbor lookups and memory-mapped files for caching
- Designed specifically for AI agent workloads that need to traverse millions of relationships across live databases
- Currently at version 0.1.0 (early alpha) with production use discouraged, supporting PostgreSQL versions 13-18