Overview

NeuG provides a set of tools for moving data in and out of your graph database.

Architecture

The data ingestion pipeline in NeuG is built around a layered design:


External Files (CSV, JSON, Parquet, ...)
        │
        ▼
   ┌───────────┐    Schema inference, relational operations
   │ LOAD FROM │    (projection, filtering, type casting, aggregation, ...)
   └────┬──────┘
        │  Unified internal format
        ▼
   ┌───────────┐    Persist or temporarily import into graph storage
   │ COPY FROM │    (auto_detect=true by default since v0.1.2)
   │ COPY TEMP │    (always auto-infers schema, session-scoped)
   └───────────┘

LOAD FROM is the foundation of data ingestion. It reads external files, automatically infers the schema, and produces a temporary result set. You can apply relational operations — such as column projection, type casting, filtering, and aggregation — directly on the loaded data.

COPY FROM builds on top of LOAD FROM. It takes the result of a LOAD FROM operation and persists it into graph storage. Because COPY FROM uses LOAD FROM internally, any file format supported by LOAD FROM is automatically available for COPY FROM as well.

COPY TO works in the opposite direction — it exports query results to external file formats.

Embedded Mode Only

Important: LOAD FROM, COPY FROM, COPY TEMP, and COPY TO are supported only in embedded mode. They are not available when NeuG is running as a service (HTTP/TP mode). This is a current limitation; support for bulk loading in service mode is planned for a future release.

Bulk file I/O operations (LOAD FROM, COPY FROM, COPY TO) involve reading or writing large files, which are long-running, I/O-intensive operations that would block the transaction processing pipeline. For this reason, they are restricted to embedded mode. Once the service is running, you can still insert individual records via CREATE statements, modify data with MERGE/SET/DELETE, and manage schema with CREATE/DROP/ALTER TABLE.

Supported Formats

Format	Supported	Availability
CSV	✅	Built-in
JSON / JSONL	✅	Built-in (since v0.1.2)
Parquet	✅	ViaParquet Extension

Version Note: In NeuG < 0.1.2, JSON/JSONL support was provided via the JSON Extension and required INSTALL json; LOAD json; before use. Since NeuG >= 0.1.2, JSON/JSONL is a built-in feature — no extension installation or loading is needed.

Note: As new format extensions are developed, both LOAD FROM and COPY FROM gain support automatically. See the Extensions page for details.

What’s Next

LOAD FROM — Read external files into temporary tables with relational operations

COPY FROM / COPY TEMP — Persist or temporarily import external data into graph storage

COPY TO — Export query results to external files