PATROAM is an always-on personal agent that runs on your machine. It blends a knowledge graph, a vector store, and retrieval-augmented generation so it can reason over your context — news, notes, stocks, media — and act through skills.
What it does
Reads and summarizes news on a schedule
Answers questions over a personal knowledge graph + vector DB
Exposes a minimal orb UI that stays out of the way
Runs skills: media control, notifications, ad-stats fetching, and more
Tech Stack
PythonLangGraphRAGVector DBWebView
Gallery
Challenges
Keeping retrieval fast and relevant while running entirely locally, and designing a graph + vector hybrid that stays consistent as memory grows.
Lessons Learned
Local-first changes every tradeoff. Latency, model size, and storage all push you toward smaller, sharper components and aggressive caching.
Future Improvements
Tighter multi-agent orchestration, a Slack bridge, and a richer skill marketplace.