Deploy and Host FalkorDB Cluster on Railway

FalkorDB is a high-performance graph database purpose-built for GraphRAG, knowledge graphs, and AI/ML applications. It uses GraphBLAS for sparse adjacency matrix graph representation, delivering low-latency Cypher queries on highly connected data.

This template deploys a multi-node FalkorDB cluster with:

Multiple primary nodes (each owning a hash-slot range) and replica nodes for failover
A FalkorDB Browser service for interactive graph exploration
Persistent Railway volumes per node so data survives restarts
Auto-generated connection variables (FALKORDB_PRIVATE_URL, FALKORDB_PUBLIC_URL, FALKORDB_PASSWORD, FALKORDB_HOST, FALKORDB_PORT)

Each graph lives on one shard. Clustering helps when you have many graphs or high write throughput distributed across graph names; it does not partition a single large graph across shards.

Common Use Cases

GraphRAG for LLM Applications — Ground LLM responses in a structured knowledge graph to reduce hallucinations and improve retrieval accuracy
Agentic Memory at Scale — Give AI agents persistent, queryable memory across sessions with high-availability guarantees
Recommendation Engines — Model user–item relationships and traverse them for real-time personalized recommendations across large datasets
Fraud Detection — Detect suspicious transaction patterns through multi-hop graph traversal with replica reads spreading query load
Multi-Tenant Knowledge Graphs — Isolate tenants into separate named graphs while sharing cluster infrastructure

Connecting to the Cluster

After deployment, Railway injects connection variables into your services. Use FALKORDB_PRIVATE_URL for services inside the same Railway project (traffic stays on Railway's private network). Use FALKORDB_PUBLIC_URL for external or local access (requires TCP Proxy enabled on the FalkorDB service).

Always use cluster-aware clients — the cluster may redirect commands with MOVED responses when a graph key lives on a different node. Both FalkorDB.from_url() (Python) and the TypeScript SDK auto-detect cluster mode at connection time.

Python

import os
from falkordb import FalkorDB

# from_url() probes INFO server and auto-detects cluster mode — no extra flag needed
# Swap FALKORDB_PRIVATE_URL for FALKORDB_PUBLIC_URL for external/local access
db = FalkorDB.from_url(os.environ["FALKORDB_PRIVATE_URL"])
graph = db.select_graph("MyGraph")

graph.query("CREATE (:Person {name: 'Alice', role: 'Engineer'})")
result = graph.query(
    "MATCH (p:Person) WHERE p.role = $role RETURN p.name",
    {"role": "Engineer"}
)

Node.js / TypeScript

import { FalkorDB } from 'falkordb';

const db = await FalkorDB.connect({
  socket: {
    host: process.env.FALKORDB_HOST,
    port: Number(process.env.FALKORDB_PORT)
  },
  password: process.env.FALKORDB_PASSWORD
});

const graph = db.selectGraph('MyGraph');
await graph.query(`CREATE (:Person {name: 'Alice', role: 'Engineer'})`);

const result = await graph.query(
  `MATCH (p:Person) WHERE p.role = $role RETURN p.name`,
  { params: { role: 'Engineer' } }
);

db.close();

redis-cli (quick test, cluster mode)

# From a service inside Railway (private endpoint)
redis-cli -c -u "$FALKORDB_PRIVATE_URL" GRAPH.QUERY MyGraph "RETURN 'connected' AS status"

# From your local machine or external client (public endpoint)
redis-cli -c -u "$FALKORDB_PUBLIC_URL" GRAPH.QUERY MyGraph "RETURN 'connected' AS status"

The -c flag enables cluster-follow mode so redis-cli automatically follows MOVED redirects.

Browser Interface

The FalkorDB Browser is included in this deployment. After deploying, click the public domain in your Railway service settings to open it. Use FALKORDB_PASSWORD from the Variables tab to log in.

Data Persistence

Each cluster node has a dedicated Railway volume. Data survives restarts and redeployments. Replica nodes provide an additional copy of your data for failover. For production workloads, configure Railway volume backups and review FalkorDB durability options.

Deploy FalkorDB Cluster