A managed graph database, GraphRAG retrieval, and the step from retrievers to agents.
Knowledge, agent brain, context memory, semantic bridge, and graph analytics, in one graph stack.
Traditional databases struggle with connected data:
Graphs excel at relationship-heavy queries that would require dozens of JOINs in SQL.
Neo4j provides unique capabilities for building AI applications:
GraphRAG Foundation:
Production-Ready:
Retrievers and graph access become tools. Three reference agents show the patterns:
fleet-agent-demo/agent: a Strands ReAct agent answering natural-language questions over the aviation fleet graph.
fleet-agent-demo/agent
graph_query
vector_search
neo4j-graphrag
BedrockAgentCoreApp
neo4j-agentcore-mcp-server: the Neo4j MCP server deployed so any agent can reach the graph.
neo4j-agentcore-mcp-server
get-schema
read-cypher
An agent that forgets every session starts from zero each time. Memory makes it persistent and personal.
Stored in Neo4j, memory is a graph: entities resolve and connect instead of piling up as isolated chunks.
The Neo4j Labs agent-memory library backs agent memory with a graph.
user_identifier=
neo4j-agentcore-agents/finance-agent wires memory in as Strands tools.
neo4j-agentcore-agents/finance-agent
core/memory.py
search_context
add_memory
get_user_preferences
get_entity_graph
:User
The result: agents that reason over connected data, grounded in the graph.