bjb2/hermes-llm-wiki-provider

hermes-llm-wiki-provider is a memory plugin for hermes-agent, an AI assistant framework. It gives the agent a form of long-term memory by maintaining a structured wiki on disk, a collection of interconnected markdown pages where each page covers a specific concept or entity that has come up in conversations. The approach keeps raw conversation history in an append-only source folder, then uses a background process to periodically have a language model distill that material into cleaner, curated wiki pages with cross-links between related concepts. This means each new conversation turn is saved instantly with no AI involvement, while the heavier work of organizing knowledge happens separately in the background. When the agent needs to recall something, it searches the wiki using a keyword index and expands outward through link relationships between pages, without an AI call for retrieval. The agent can also explicitly ask the memory system to store a specific fact, which gets folded into the wiki on the next compilation pass. The plugin exposes four tools to the agent: searching the wiki by query, reading a specific page in full, exploring the links around a page, and explicitly storing a fact. Configuration lets you point the compilation step at any language model with an OpenAI-compatible API, keeping it independent of the main agent's model, so compilation can run on a cheaper or separate endpoint. You would use this if you are building or extending a hermes-agent setup and want it to accumulate knowledge across sessions rather than starting fresh each time. The plugin is written in Python, has an offline test suite that runs without a live language model, and is released under the MIT license.