OpenJarvis

Why we included this project

Most personal AI assistants still send your data to a server for every reply. OpenJarvis is a framework for building those assistants to run locally instead, on your own hardware. It provides reusable primitives for composing on-device agents, plus ready-made presets covering daily briefings, multi-hop research, scheduled monitoring, and a code assistant that can execute Python. The evaluation harness is the interesting part: it treats energy, latency, FLOPs, and dollar cost as real constraints alongside accuracy, and a learning loop improves models from local trace data. That combination makes it useful as a research platform and as a base for privacy-sensitive assistant products on laptops and edge devices.