Show HN: AgentOS- a memory system for AI agents that learns what it doesn't know

Happy to answer questions on any of the technical decisions.

The one I'm most uncertain about: whether SQLite WAL is the right store for memory at this scale, or if a proper vector DB starts making sense once you have thousands of chunks. Right now retrieval is keyword-scored, not semantic — fast, but misses synonyms and paraphrased queries.

If anyone has run into this tradeoff, I'd love to hear how you approached it.

GitHub: https://github.com/ajstars1/agent-os