neboy72/hermes-nexus-memory

Hermes Nexus Memory is a Python plugin that gives Hermes Agent persistent memory using Qdrant, a vector database, so the agent no longer forgets everything at the end of a session. It lets an AI agent search past facts, decisions, and patterns by meaning rather than exact keyword matching, and that stored information survives restarts, model switches, and gateway restarts. The plugin exposes three tools: one to search memories semantically, one to save a new memory with a category and source, and one to delete a specific memory by its ID. It works with three different embedding providers, which are the underlying systems that turn text into the numeric vectors used for semantic search. The default is sentence-transformers, which runs locally in Python and needs no account or API key. A second option runs through Ollama as a local service. The third, Voyage, is a paid cloud API that offers the highest quality results but requires signing up for an API key. Setup is done with a single install script, followed by restarting the Hermes gateway, or through a built-in setup wizard that walks you through picking an embedding provider. The system requires Python 3.11 or newer and a Qdrant server running locally on the default port. The author positions this as a smaller and simpler alternative to other memory plugins, at roughly 400 lines of Python code, with local operation by default and no mandatory external API dependency, compared to some alternatives that require thousands of lines of code or a mandatory cloud account. The project includes a troubleshooting table covering common issues like missing tools, a Qdrant server that is not running, or lost memories after switching embedding providers. It is licensed under MIT.