agiresearch/a-mem — reverse-engineered prompt

Build me a Python library for agent memory that I can plug into an LLM agent.

I want it to let me save notes or experiences, then automatically organize them with useful tags, keywords, context, and links to related memories. It should support adding, reading, searching, updating, and deleting memories. Search should feel smart, so if I ask about an idea it can find related memories even when the wording is different.

The memory system should evolve over time. When I add or update something, it should look at older memories, find connections, improve the metadata, and keep a network of related notes, kind of like a Zettelkasten.

Please make it work with an embedding based vector store for persistence and semantic search, and let me use either OpenAI or a local Ollama model for the LLM parts. Include simple examples, basic tests, and clear setup instructions so I can install it and try saving and searching memories right away.