Hexus
A Postgres-backed vector memory layer for AI agents, using a local BERT model for embedding search with no LLM calls in the hot path, distributed as a plugin for the hermes-agent framework and as a standalone MCP server.
Hexus provides scoped, searchable vector memory for teams of cooperating AI agents, built on Postgres and pgvector rather than a separate hosted vector database. Each agent or workflow gets its own isolated memory space, called a theme, so one agent's notes do not leak into another's; embedding search runs entirely through a local BERT model with no LLM call in the retrieval path, aimed at lower latency and full data privacy.
It installs as a plugin for the hermes-agent framework, either through pip or by dropping the module directly into a Hermes plugins directory for automatic discovery, and requires two configuration blocks in the Hermes config: one enabling the memory plugin system, and one pointing it at Hexus with a Postgres connection string. It also runs as a standalone Model Context Protocol server usable by Claude Desktop, Cursor or other MCP clients independent of Hermes.
It positions itself against cloud memory providers that route every read and write through an LLM, offering a local storage backbone instead.