ardhaecosystem/synapse — reverse-engineered prompt

Build me a self hosted memory system for an AI agent, something like a synthetic hippocampus instead of a basic chat log. I want it to run locally with Docker for the graph database, keep conversation data private, and plug into a Hermes style agent as its memory provider.

The memory should automatically pull out people, places, facts, and relationships from conversations, remember when things were true, and answer questions about the past as well as the present. It should rank what matters, let unimportant memories fade over time, notice contradictions or genuinely new information, and strengthen memories when they get recalled. I also want a background consolidation step that acts like sleep, plus a way for the agent to explicitly save something forever if it is important.

Please make it work with any OpenAI compatible model endpoint, including local ones if possible, and give me a simple setup flow, sensible defaults, and a small demo showing memory being written and queried over time. If you need details, look up current docs online.