LongMemory

OpenMemory is an open source long term memory system for AI agents and assistants. Rather than storing flat embeddings, it encodes conversations into five sectors: factual, emotional, temporal, relational and behavioral. Memories decay along configurable curves with per sector slopes and floors, and reinforcement pulses restore strength when conversations or tool outcomes signal that context still matters. Each reinforcement links back to its trigger, so an agent can explain why a memory stayed active. Retrieval uses graph based, multi hop navigation with explainable paths, and ingestion handles multiple file formats. It is aimed at developers building agents, assistants and knowledge graph applications who want to run memory themselves. The vendor lists self hosting, local embeddings and full data ownership, and positions it as an alternative to hosted memory services such as Supermemory and Mem0. Deployment is described as taking minutes, with documentation linked from the home page. The current engine generation is called HMD v2. No plans or subscription prices are published. The site's cost figures are self reported estimates for running the software: about $0.35 per million tokens of hosted embeddings, and an expected $5 to $8 per month at 100,000 memories. These are the vendor's own comparisons, not prices charged for a service.