agentforge-memory-neo4j

agentforge-memory-neo4j

Neo4j-backed MemoryStore and GraphStore for the AgentForge framework.

What this is

• Neo4jMemoryStore — MemoryStore (claim audit log) over Neo4j. Claims map to (:Claim {…}) nodes; queries become parameterised Cypher.
• Neo4jGraphStore — GraphStore (knowledge graph traversal) over Neo4j. Nodes and edges map directly to property graph nodes and relationships. match() and traverse() compile to Cypher.

Both pass the locked agentforge_core.testing.run_*_conformance suites — drop-in alternatives to agentforge-memory-sqlite when you need real graph semantics, multi-writer concurrency, or production operability (clustering, fine-grained auth, query planner).

Usage

from agentforge_memory_neo4j import Neo4jGraphStore

async with Neo4jGraphStore.from_url(
    "bolt://localhost:7687",
    auth=("neo4j", "password"),
    database="neo4j",
) as store:
    await store.add_node(GraphNode(id="paper:1", labels=("Doc",)))
    ...

Local development

docker compose -f docker-compose.dev.yml up -d
RUN_LIVE_NEO4J=1 NEO4J_URL=bolt://localhost:7687 \
  NEO4J_USER=neo4j NEO4J_PASSWORD=test \
  uv run pytest packages/agentforge-memory-neo4j/tests/integration -v

Capabilities

{"transactions", "cypher", "fulltext"} — Neo4j 5.x ships native fulltext indexes; transactions are first-class (every write goes through session.execute_write).

Vector search is not declared: Neo4j 5.x has vector indexes but adopting them is tracked separately. Pair Neo4j with agentforge-memory-postgres (pgvector) or agentforge-memory-sqlite for embeddings.