Overview

AkasicDB is an integrated graph, vector, and relational DBMS for enterprise AI. It processes graph, vector, and relational data together within a single database, enabling AI applications to be built without a separate AI-specific database or complex data pipelines.

AkasicDB is compatible with the existing PostgreSQL ecosystem and provides libraries and tools for AI development. It can also build knowledge graphs from existing table data within a single database.

Key Features

Integrated Search and Query Processing

AkasicDB executes vector search, graph traversal, and SQL predicates together within a single query.

Instead of combining separate query results at the application layer, users can perform unified search over vector, graph, JSON, relational, and text data in a single engine. This enables AI agents and LLMs to use enterprise data safely and consistently.

Integration with the Existing Ecosystem

AkasicDB supports LangChain integration for agentic AI applications. PostgreSQL compatibility allows AkasicDB to work with existing tools, libraries, and operational environments.

Knowledge Graph Extraction from Relational Data

AkasicDB enables users to derive knowledge graphs directly from relational tables using SQL. It lets users model the structure and relationships already present in relational data as graphs, without requiring data conversion or migration. This reduces the time and cost needed for initial deployment.

Core Technologies

Triple Storage

AkasicDB’s triple storage is an integrated storage architecture designed to process graph, vector, and relational workloads within a single logical database. AkasicDB uses workload-specific execution paths for graph traversal, vector search, and relational lookup, while exposing them through a single, consistent DBMS interface.

This reduces the burden of data movement and duplicate management while providing the performance and consistency required for various AI workloads.

Traversal-Join-Similarity Operator

In traditional approaches that split complex queries across multiple databases, developers must decide the execution order for vector search, graph traversal, and SQL lookup themselves, which can lead to inefficient data access.

Based on the Traversal-Join-Similarity operator, AkasicDB optimizes vector search, graph traversal, and SQL lookup as a single query plan. This allows AkasicDB to generate efficient query plans without requiring changes to application logic.

Comparison with other PostgreSQL Extensions

There have been continuous efforts to extend relational DBMSs with graph query or vector search capabilities, such as Apache AGE and pgvector. However, simply adding graph or vector capabilities on top of a relational data model and optimization structure can limit query-performance optimization when data volumes are large or queries are complex.

AkasicDB, however, handles graph, vector, and relational workloads within an integrated execution model, with an architecture and optimization framework beyond conventional extension-based approaches.

Comparison with Apache AGE

Feature	AkasicDB	Apache AGE
Graph storage model	Stores graph topology in a dedicated graph format	Represents graphs using relational tables
Graph traversal model	Performs graph traversal with native traversal operations	Performs graph traversal through relational join operations
Graph definition flexibility	Provides high flexibility through SQL-based graph definitions	—

Comparison with pgvector

Feature	AkasicDB	pgvector
Dimension limit	Up to 65,535	Up to 2,000
Index storage model	Uses dedicated disk and memory structures for index management	Uses PostgreSQL’s built-in storage
SIMD optimization	Selects CPU-specific implementations dynamically at runtime	Uses static CPU dispatch on Linux x86-64
WAL support	Full WAL support for data; index WAL support in progress	Full WAL support for data and indexes
Filtered vector search	Fully supported through the VBASE approach	May return fewer results when filters are applied