Distributed long-term memory. Milvus semantic search, episodic/semantic hierarchy, and millisecond retrieval.
Every time an agent restarts, it loses all memory of previous conversations. Traditional vector databases store embeddings but not relationships, hierarchies, or temporal evolution of knowledge. You need memory that works like the human brain.
World-class vector search engine. Millions of documents indexed with sub-100ms latency. Hybrid: vector + full-text + structured filters.
Memory organized in episodes, scenes, and details. Like the human brain. Remember not just what happened, but when, where, and in what order.
Abstract concepts, relationships, and general knowledge. Agents reason about ideas, not just concrete facts. Automatic inference and generalization.
Automatic sharding, replication, and time-based partitioning. From gigabytes to petabytes without reconfiguration. High availability with automatic failover.
Django with Milvus for vectors, PostgreSQL for metadata, and Redis for cache. Microservices architecture.
Backend
Django 4.2+
Vectors
Milvus 2.3+
Metadata
PostgreSQL 15+
Cache
Redis 7+
Protocol
gRPC + REST
Scaling
Auto-sharding
SomaFractalMemory persists all knowledge of the SomaTech ecosystem. Every conversation, every document, every insight is stored forever.
Apache 2.0 License
Store up to 1 million vectors free. Complete Milvus with hybrid search. No collection limits.
On request
Unlimited vectors, multi-node clustering, automatic backup, and support for custom embeddings.
Give long-term memory to your AI agents with SomaFractalMemory. Never forget a conversation again.
No commitments. No credit card. Results in 48 hours.
