mem0-falkordb 0.3.0

FalkorDB graph store plugin for Mem0

mem0-falkordb

FalkorDB graph store plugin for Mem0. Adds FalkorDB as a graph memory backend without modifying any Mem0 source code.

Installation

pip install mem0-falkordb

You also need Mem0 installed separately:

pip install mem0ai

Quick Start

from mem0_falkordb import register
register()

from mem0 import Memory

config = {
    "graph_store": {
        "provider": "falkordb",
        "config": {
            "host": "localhost",
            "port": 6379,
            "database": "mem0",
        },
    },
    # Add your LLM and embedder config as usual
    "llm": {
        "provider": "openai",
        "config": {"model": "gpt-4o-mini"},
    },
}

m = Memory.from_config(config)
m.add("I love pizza", user_id="alice")
results = m.search("what does alice like?", user_id="alice")

Configuration

Parameter	Type	Default	Description
host	str	localhost	FalkorDB server host
port	int	6379	FalkorDB server port
database	str	mem0	Graph name prefix (each user gets {database}_{user_id})
username	str	None	Authentication username (optional)
password	str	None	Authentication password (optional)
base_label	bool	True	Use __Entity__ base label

Per-User Graph Isolation

Each user automatically gets their own isolated FalkorDB graph (e.g. mem0_alice, mem0_bob). This leverages FalkorDB's native multi-graph support and provides:

• Natural data isolation — no user_id filtering needed in Cypher queries
• Simpler, faster queries — no WHERE clauses on user_id
• Easy cleanup — delete_all simply drops the user's graph

Running FalkorDB

Using Docker:

docker run --rm -p 6379:6379 falkordb/falkordb

How It Works

This plugin uses Python's runtime patching to register FalkorDB into Mem0's existing factory system:

1. GraphStoreFactory.provider_to_class gets a new "falkordb" entry
2. GraphStoreConfig is patched to accept FalkorDBConfig
3. A MemoryGraph class translates Mem0's graph operations to FalkorDB-compatible Cypher

Key Cypher Translations

Neo4j	FalkorDB
elementId(n)	id(n)
vector.similarity.cosine()	db.idx.vector.queryNodes() procedure
db.create.setNodeVectorProperty()	SET n.embedding = vecf32($vec)
CALL { ... UNION ... } subqueries	Separate outgoing + incoming queries

Development

git clone <repo>
cd mem0-falkordb
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest

License

Apache-2.0