OpenJarvis is a framework for building local-first personal AI agents that run on personal devices without routing intelligence through cloud APIs.
The project addresses the gap between growing demand for personal AI assistants and the practical challenges of running them locally. Rather than depending on cloud services, OpenJarvis provides a software stack designed around three core principles: shared primitives for constructing on-device agents, evaluations that treat energy consumption, computational cost, latency, and dollar cost as first-class constraints alongside accuracy, and a learning loop that improves models using local trace data. The framework aims to make local-first operation the default, calling cloud services only when necessary.
OpenJarvis suits developers and researchers building personal AI systems who want to minimize cloud dependency and control their inference infrastructure. The project positions itself as both a research platform and a production foundation for local AI, comparable in scope to foundational tools in the machine learning ecosystem. It is particularly relevant for those working with resource-constrained devices or seeking to optimize for energy efficiency and latency alongside model performance. The installation process is streamlined across multiple platforms, with a single command handling dependency setup including the Python environment, Ollama, and an initial model.
The project maintains active development with regular updates to its documentation, leaderboard for evaluating local models, and a published roadmap for future work. The codebase is written in Python and requires Python 3.10 or later. The team publishes research alongside the software, including a paper describing the framework's design and findings on local model capabilities. Community engagement occurs through documented Discord and social media channels.
