Migration Guide

If you've worked with Cypher-style or other graph query systems before, most of what you know transfers. ArcFlow's WorldCypher is Cypher-inspired; your existing MATCH, CREATE, MERGE, and WHERE clauses usually carry over directly. What changes: the deployment model (in-process or lightweight server), built-in GPU-accelerated algorithms, confidence scoring, temporal queries, and vector search with no add-ons.

This guide covers what stays the same, what's different, and how to move your data.

Architecture Differences#

ArcFlow can run as an in-process engine with zero network overhead, or as a standalone server. The in-process mode eliminates serialization and connection management entirely.

GQL Compatibility#

If you're coming from a Cypher-family system, most of your queries port directly. If you're coming from a different graph query model, expect a query rewrite, but the graph modeling concepts still transfer.

Fully Compatible (GQL family)#

These Cypher features work identically in WorldCypher:

Statements: CREATE, MATCH, OPTIONAL MATCH, WHERE, RETURN, ORDER BY, SKIP, LIMIT, DELETE, DETACH DELETE, SET, REMOVE, MERGE, WITH, UNION, UNWIND, CASE WHEN

Predicates: AND, OR, NOT, XOR, =, <>, <, >, <=, >=, IS NULL, IS NOT NULL, IN, CONTAINS, STARTS WITH, ENDS WITH, EXISTS

Functions: id(), labels(), type(), count(), sum(), avg(), min(), max(), collect(), coalesce(), toString(), toInteger(), toFloat(), toLower(), toUpper(), trim(), substring(), left(), right(), replace(), split(), reverse()

Patterns: Node patterns (n:Label {prop: val}), relationship patterns -[:TYPE]->, variable-length paths [*1..5], shortestPath()

Schema: CREATE INDEX, CREATE CONSTRAINT (unique), DROP INDEX, DROP CONSTRAINT

WorldCypher Extensions#

Features in WorldCypher that go beyond standard GQL — ArcFlow-specific capabilities:

Not Yet Supported#

Standard GQL features or ecosystem-specific extensions not yet available in WorldCypher:

Performance Snapshot#

Recent ArcFlow benchmark runs on the current engine show strong single-node throughput in the embedded path:

For the exact benchmark methodology and the latest corrected report, see the engine benchmark artifacts in the main ArcFlow repository.

Migration Steps#

Step 1: Export from Your Current Database#

Export your data as Cypher statements, CSV, or JSON from your current stack. The exact command depends on your existing tooling, but the migration target is the same: node rows, relationship rows, and any schema definitions you want to recreate.

-- Most GQL-compatible databases support exporting as CREATE statements
-- Adapt vendor-specific export commands to your setup

Step 2: Adapt the Export#

WorldCypher accepts most Cypher CREATE statements directly. Key adaptations:

1. Remove environment-specific syntax (proprietary ID functions, internal properties)
2. Replace FOREACH with UNWIND
3. Remove CALL {} IN TRANSACTIONS wrappers
4. Convert non-Cypher query definitions into Cypher-style mutations where needed
5. Keep CREATE, MERGE, and SET statements as-is

Step 3: Import into ArcFlow#

# Start ArcFlow CLI
arcflow --data-dir ./my-graph
 
# Import CSV files
:import csv /tmp/nodes.csv

Or use CREATE statements directly:

-- Nodes import directly
CREATE (n:Person {name: 'Alice', age: 30})
CREATE (n:Person {name: 'Bob', age: 25})

-- Relationships
MATCH (a:Person {name: 'Alice'}), (b:Person {name: 'Bob'})
CREATE (a)-[:KNOWS {since: 2020}]->(b)

Step 4: Recreate Indexes#

CREATE INDEX ON :Person(name)
CREATE INDEX ON :Person(email)
CREATE CONSTRAINT ON (p:Person) ASSERT p.email IS UNIQUE

Step 5: Verify#

-- Check counts match
CALL db.stats()

-- Check schema
CALL db.schema()

-- Run diagnostics
CALL db.doctor()

Graph Analytics: Built In#

ArcFlow ships graph algorithms directly in the engine. No projection step, no separate installation.

Key difference: no projection step. With most graph analytics libraries, you first create an in-memory graph projection, run the algorithm, then optionally write results back. In ArcFlow, algorithms run directly on the live graph store with zero copy.

-- Traditional analytics workflow (3 steps):
-- 1. Project graph into analytics engine
-- 2. Run algorithm on projection
-- 3. Drop projection
 
-- ArcFlow (1 step):
CALL algo.pageRank()

When to Use Which#

Use ArcFlow when:

• Embedding a graph engine in your application (no server management)
• You need GPU-accelerated graph algorithms
• You want built-in vector search + graph context (GraphRAG)
• Temporal queries and confidence scoring matter for your domain
• You want a single binary with no runtime dependency

Use a server-oriented graph platform when:

• You need multi-user concurrent access at scale
• Full ACID transactions are a hard requirement
• Your graph exceeds available RAM
• You need a large plugin ecosystem
• You need enterprise features (RBAC, causal clustering, CDC)

See Also#

• QuickStart -- get running in 5 minutes
• Built-in Functions -- all 93 functions
• Procedures -- 100+ procedures
• RAG Pipeline -- building RAG with ArcFlow