WHY NEUROSYMBOLIC

Stop syncing two databases. Start reasoning.

Today teams duct-tape a vector database to a knowledge graph and write glue to keep them in sync. Oxid collapses that into one engine, one query language, one identity space: where logic filters during the vector search, not after.

See how it works Read the docs

Neural+Symbolic=Oxid

THE STATUS QUO

Two stores. Two truths. A layer of glue holding them together.

VECTOR DB

Vector DB

Embeddings & similarity search

sync glue

KNOWLEDGE GRAPH

Knowledge Graph

Entities & explicit relations

Embeddings drift from the graph. Identity gets duplicated across systems. The glue turns brittle with every schema change, and every query ends in fragile post-filtering, accuracy quietly falling through the cracks.

Oxidone engine · one query · one identity space

WHO IT'S FOR

Built for the people wiring intelligence into production.

Backend, ML & platform engineers

Replace a brittle vector-DB-plus-graph stack with one engine you can actually operate: ACID, durable, one query language.

AI agent developers

Give agents memory that reasons: retrieval that respects ontology, identity, and provenance instead of fuzzy nearest-neighbors.

Founders & technical teams

Ship intelligent features without standing up and syncing two databases. One dependency, one mental model, faster to ship.

Applied-AI & research teams

Experiment on reasoning-aware retrieval with structural embeddings and logic: all in-engine, no glue, fully reproducible.

WHAT YOU CAN BUILD

Use cases that need logic and similarity at once.

Reasoning-aware RAG

Retrieval that filters by ontology and constraints during the vector search: so context is relevant and logically valid, not just semantically close.

RAGRetrieval

Durable agent memory

Long-lived memory agents can reason over: entities with identity, relations, and history. Survives restarts with ACID guarantees.

AgentsMemory

Ontology-backed knowledge bases

Model your domain as a real graph with typed relations and inference, then query it with vectors and logic in one language.

Knowledge graphOntology

Provenance-tracked retrieval

Every retrieved fact carries where it came from and when: auditable, explainable answers instead of opaque nearest-neighbors.

ProvenanceAudit

HOW IT COMPARES

The patchwork, side by side.

Great databases, each covering half the picture. Vector engines can't reason, relational and document stores can't search by meaning, and even multi-model engines stop short of native, unified ontology, inference, and provenance.

Capability	Oxid	Weaviate	Qdrant	Postgres	MongoDB	RavenDB	SurrealDB
FOUNDATIONS
Vector similarity search
Native graph relations & traversal
Metadata-filtered vector search
ACID transactions
One query language for vectors, graph & logic
THE NEUROSYMBOLIC LAYER
Ontology & type hierarchies
In-engine rules & logical inference
Logic constraints applied during the vector search
Structural graph embeddings in-engine
First-class provenance on every fact
Temporal facts & history
One identity space across vectors, graph & logic

Built inPartial: via extension, add-on layer, or limited scopeNot available

Postgres adds vectors through the pgvector extension. MongoDB now runs vector search as a GA feature on both Atlas and self-managed editions, in a separate but query-integrated mongot process. Weaviate supports reference-path navigation, but not a full graph-query language. RavenDB keeps stored document-version history through revisions, not a semantic bitemporal fact model. SurrealDB natively spans document, graph, vector, and time-travel, and its first-party Spectron layer adds controlled ontology grounding, provenance, and temporal belief; formal inference and full ontology reasoning still aren't core-engine features.

HONEST SCOPE

What Oxid is not.

A new category needs clear edges. Oxid is a neurosymbolic reasoning engine, not a drop-in for your system of record.

Not a Postgres replacement

Keep Postgres for transactional system-of-record workloads. Oxid sits beside it as the reasoning layer.

Not a general-purpose RDBMS

No sprawling SQL surface or OLAP warehouse ambitions. It's built for retrieval and reasoning, not reporting.

Not 'just another vector DB'

Vectors are half the story. Logic, identity, and provenance are first-class, not a bolted-on filter.

What it is: a neurosymbolic reasoning engine (vectors, logic, identity, and provenance sharing one mind).

ONE ENGINE · ONE QUERY · ONE MIND

Stop maintaining the glue. Start reasoning over your data.

Collapse the vector DB, the knowledge graph, and the sync layer into one neurosymbolic engine. See what your retrieval can do when logic and similarity finally share a mind.

Start building Read the docs Talk to us →