neo4jev

What neo4jev does

neo4jev demonstrates graph navigation where each hop is a typed Jev decision rather than a generative Cypher or natural-language guess. At a visited node, outgoing relationships become Choice options (type, properties, target label); a Noul “has the goal been reached?” question rides in the same system_one call so every hop is one round-trip. Probabilities feed a beam search ranked by summed log-probabilities, and results render with neo4j-viz beside the surrounding neighbourhood. The library is schema-agnostic: labels and relationship types are discovered live, with Jupyter notebooks and a Streamlit app layered on a shared Python core under src/neo4jev. Default demos target Neo4j’s public companies knowledge graph, but any instance works via environment config once TYPESAFE_API_KEY and Neo4j credentials are set. Goal modes cover free-text goals, explicit target nodes, and path-intent patterns that return several high-scoring beams. This page is for builders who already store knowledge in Neo4j and want Jev to pick the next edge under calibrated uncertainty instead of inventing paths in prose.

Where it fits in the Jev stack

• Exploratory multi-hop search over company or domain knowledge graphs
• Teaching beam search with calibrated edge probabilities
• Streamlit UIs that visualize Jev-chosen paths beside neighbourhood context

Browse more in Integrations, or return to the Jev AI tools directory.