A Stanford framework for AI that lives on your laptop, not their cloud

What it does

OpenJarvis is a local-first framework for building personal AI agents on your own hardware. Developed at Stanford’s Scaling Intelligence Lab, it ships with eight built-in agents—including a morning briefing generator, a deep-research assistant, and persistent monitors—and a CLI that wraps Ollama and local models. The stack treats energy, FLOPs, latency, and dollar cost as first-class constraints rather than optimization afterthoughts.

The interesting bit

The project is backed by research showing local models already handle 88.7% of single-turn chat and reasoning queries, with intelligence efficiency improving 5.3× from 2023 to 2025. Instead of treating on-device inference as a fallback, OpenJarvis centers on it: a learning loop refines models using local trace data, and a skills ecosystem—importable from Hermes Agent or OpenClaw’s 13,700 community entries—teaches agents new tricks without adding cloud dependencies.

Key highlights

• Local-first by default: agents run on-device and call the cloud only when necessary
• Eight built-in agents across scheduled, on-demand, and continuous execution modes
• Skills follow the open agentskills.io standard and can be imported from public catalogs
• Evaluations weigh energy, latency, FLOPs, and cost alongside raw accuracy
• Ships as a cross-platform CLI, desktop GUI, and background Rust extension

Verdict

Worth exploring if you’re building agents where privacy, offline resilience, or compute costs matter; skip it if you just want a drop-in cloud chatbot wrapper.