Multi-Tenancy

Graphmind supports multiple isolated graph namespaces (tenants) within a single server instance. Each graph has its own nodes, edges, indexes, and data -- completely separated from other graphs.

How It Works

Every query targets a specific graph by name. The default graph is called default. Graphs are created automatically on first write -- no explicit setup needed.

Creating and Using Graphs

HTTP API

Specify the graph name in the graph field:

# Write to a graph named "analytics"
curl -X POST http://localhost:8080/api/query \
  -H 'Content-Type: application/json' \
  -d '{"query": "CREATE (n:User {name: \"Alice\"})", "graph": "analytics"}'

# Write to a different graph
curl -X POST http://localhost:8080/api/query \
  -H 'Content-Type: application/json' \
  -d '{"query": "CREATE (n:User {name: \"Bob\"})", "graph": "marketing"}'

If you omit the graph field, it defaults to "default".

RESP Protocol

The graph name is the first argument to GRAPH.QUERY:

redis-cli -p 6379
127.0.0.1:6379> GRAPH.QUERY analytics "CREATE (n:User {name: 'Alice'})"
127.0.0.1:6379> GRAPH.QUERY marketing "CREATE (n:User {name: 'Bob'})"

Python SDK

from graphmind import GraphmindClient

client = GraphmindClient.embedded()  # or .connect(...)

# Write to different graphs
client.query('CREATE (n:User {name: "Alice"})', graph="analytics")
client.query('CREATE (n:User {name: "Bob"})', graph="marketing")

# Queries are scoped to their graph
a = client.query_readonly("MATCH (n) RETURN count(n)", graph="analytics")
m = client.query_readonly("MATCH (n) RETURN count(n)", graph="marketing")
# a and m are independent counts

TypeScript SDK

// Pass graph name as second argument
await client.query('CREATE (n:User {name: "Alice"})', "analytics");
await client.query('CREATE (n:User {name: "Bob"})', "marketing");

Rust SDK

client.query("analytics", r#"CREATE (n:User {name: "Alice"})"#).await?;
client.query("marketing", r#"CREATE (n:User {name: "Bob"})"#).await?;

Listing Graphs

HTTP API

curl http://localhost:8080/api/graphs

{"graphs": ["default", "analytics", "marketing"]}

RESP Protocol

127.0.0.1:6379> GRAPH.LIST
1) "default"
2) "analytics"
3) "marketing"

SDKs

graphs = client.list_graphs()
# ["default", "analytics", "marketing"]

Deleting a Graph

Deletes all data (nodes, edges, indexes) in a graph namespace.

HTTP API

curl -X DELETE http://localhost:8080/api/graphs/marketing

{"status": "ok", "deleted": "marketing"}

RESP Protocol

127.0.0.1:6379> GRAPH.DELETE marketing
OK

SDKs

client.delete_graph("marketing")

Embedded Mode (Rust)

In embedded mode, you can use TenantStoreManager for direct access to isolated graph stores:

use graphmind::TenantStoreManager;

let mgr = TenantStoreManager::new();

// Stores are created automatically on first access
let production = mgr.get_store("production").await;
let staging = mgr.get_store("staging").await;

// Each store is fully isolated
{
    let mut store = production.write().await;
    engine.execute_mut("CREATE (n:User {name: 'Alice'})", &mut store, "production")?;
}

// List all tenant graphs
let graphs = mgr.list_graphs().await; // ["default", "production", "staging"]

Embedded Mode (Python)

The Python SDK supports multi-tenancy in both embedded and remote modes:

client = GraphmindClient.embedded()

# Write to isolated graphs
client.query('CREATE (n:User {name: "Alice"})', graph="production")
client.query('CREATE (n:User {name: "Bob"})', graph="staging")

# Read-only queries scoped to a graph
result = client.query_readonly("MATCH (n) RETURN n", graph="staging")

# List all graphs
graphs = client.list_graphs()  # ["default", "production", "staging"]

Isolation Guarantees

• Each graph has its own node ID space, label indexes, and property indexes
• A query on graph A cannot see or affect data in graph B
• Persistence uses RocksDB column families with tenant-prefixed keys
• The default graph exists automatically and cannot be deleted

Web Visualizer

The web UI at http://localhost:8080 includes a graph selector dropdown. Use it to switch between graphs. The graph name is included in all API calls the UI makes.

Use Cases

Pattern	Example
Per-customer isolation	tenant_acme, tenant_globex
Environment separation	production, staging, test
Domain separation	social_graph, product_catalog, fraud_detection
Temporary workspaces	import_20240315, experiment_a (delete when done)