Spector
A memory engine for AI agents written in Java, storing working, episodic, semantic, and procedural memories with decay and consolidation, and retrieving them through fused semantic and SIMD-accelerated hybrid search.
Spector is a memory engine for AI agents that models four memory tiers -- working, episodic, semantic, and procedural -- and applies decay, consolidation, and association graphs across them, linking traces by co-activation, temporal, and entity edges so recall can surface related memories, not only exact matches.
Agents connect through a built-in MCP server, a REST/gRPC API, Python and TypeScript SDKs, or direct JVM embedding. Each user, agent, or tenant gets an isolated on-disk namespace. The engine uses Java's Project Panama and the Vector API for SIMD-accelerated scoring with low garbage-collection overhead, and needs no external database.
It can run as a zero-install MCP server via npx, connecting to a local daemon or an embedded in-process kernel, or be built from source with Maven and launched as a standalone server.