Descriptions:
Fahd Mirza walks through the installation and hands-on testing of Open Jarvis, a newly released local-first personal AI framework developed by Stanford’s Hazy Research and Scaling Intelligence Lab as part of their “Intelligence per Watt” research initiative. The project is built around the principle that personal AI should run on-device by default, with no cloud dependency, and integrates natively with Ollama, vLLM, and llama.cpp.
Mirza runs the demo on an Ubuntu server with an NVIDIA RTX 3060 GPU and 48GB VRAM, using a Qwen 3.5 27B model configured with an extended 32K+ context window to support tool and function calling. He covers Open Jarvis’s five core architectural primitives — Intelligence (model selection), Engine (Ollama-backed inference), Agents (multi-step reasoning and scheduling), Memory (persistent searchable storage), and a Learning system that records interaction traces to improve routing over time. The video gives particular attention to the framework’s preset system, which bundles seven ready-made agent configurations — including morning digest, deep research, and code assistant — letting users spin up specialized workflows with a single init command.
Mirza is candid that Open Jarvis is an early-stage project with occasional rough edges, including unsolicited self-update prompts and some configuration quirks. Developers looking for a practical introduction to running private, locally-hosted AI agents with Ollama-compatible models will find this a useful first look at the framework’s current capabilities and limitations.
📺 Source: Fahd Mirza · Published June 26, 2026
🏷️ Format: Tutorial Demo
