Neo brings graph memory to companions
Neo is an open-source personal companion intelligence platform that stores long-term memory in a persistent 9-dimensional graph across goals, habits, health, relationships, work, events, emotions, preferences, and context. It combines Neo4j for relationship structure with Qdrant for vector similarity, is self-hostable via Docker Compose, and is positioned as a model-agnostic system built on LangGraph and FastAPI. The project is being tested in a real deployment with PROAS in Vienna for around-the-clock support between care visits.
Strong concept, and the graph-plus-vector split is the right instinct for memory systems that need both semantic recall and explicit relationships.
- –The 9-dimensional framing is useful because it forces memory into interpretable buckets instead of one flat embedding soup.
- –Neo4j gives the project something most “memory” products lack: durable, queryable structure for identity, relationships, and causality.
- –Qdrant handles recall well, but the hard part will be keeping salience, decay, and compaction from turning the graph into noise over time.
- –The biggest product risk is not retrieval quality, but memory governance: what gets stored, how it gets updated, and when contradictions overwrite or coexist.
- –Real-world deployment with a nonprofit is the strongest signal here, because companion memory only matters if it survives messy, longitudinal usage.
DISCOVERED
10h ago
2026-04-17
PUBLISHED
10h ago
2026-04-17
RELEVANCE
AUTHOR
Lost_Cause3655