okanfx/memory-os ? reverse-engineered prompt

Build me a local memory system for Hermes Agent that makes it stop forgetting things between sessions.

I want it to run on my machine with Docker, use Qdrant and Redis under the hood, and work with whatever model provider Hermes already supports, like OpenAI, Anthropic, OpenRouter, or Ollama. It should keep workspace notes, session history, structured facts, a searchable long term memory, and an auto updating wiki so the agent can pull back the right context when it needs it.

Please make the memory feel smart and selective, not noisy. I want it to inject only the most relevant context into the prompt, avoid repeating the same thing over and over, and include a clear instruction layer so the agent actually trusts and uses the injected memory. If anything is unclear, look up current docs online and wire it together in the simplest working way.