AgentMemory setup with Hermes Agent and Ollama

This tutorial sets up AgentMemory with Hermes Agent and Ollama to retain coding context across sessions. The presenter uses Ubuntu and reports an Nvidia RTX 5600 with 48 GB of VRAM. That is the demonstration hardware, rather than a stated minimum requirement.

Hermes connects to the local LLM through a custom direct API using Ollama's OpenAI-compatible endpoint and chat completions. The presenter then installs AgentMemory, selects the Hermes integration and chooses BM25-only mode. He explains that this mode stores observations and retrieves memories through keyword search, but skips LLM compression and consolidation. The broader architecture he describes includes semantic retrieval and a knowledge graph; those capabilities are not tested in this setup.

The configuration adds AgentMemory as both a memory provider and an MCP server in Hermes's config.yml. A browser viewer at localhost port 3113 displays saved memories. The coding assistant creates a small Python Flask project, then recalls its details after the presenter starts a new session. A follow-up request adds a PUT update endpoint.

The demonstration also exposes a viewer limitation: the memory count increases, but the memories tab does not clearly display the second entry. The presenter suggests a UI bug and discusses other model providers for semantic features. The demonstrated result is a local keyword-based memory workflow, with broader retrieval claims left unverified.