{"id":1031,"date":"2026-06-04T14:23:10","date_gmt":"2026-06-04T06:23:10","guid":{"rendered":"https:\/\/connectword.dpdns.org\/?p=1031"},"modified":"2026-06-04T14:23:10","modified_gmt":"2026-06-04T06:23:10","slug":"meet-openjarvis-a-local-first-framework-for-on-device-personal-ai-agents-with-tools-memory-and-learning","status":"publish","type":"post","link":"https:\/\/connectword.dpdns.org\/?p=1031","title":{"rendered":"Meet OpenJarvis: A Local-First Framework for On-Device Personal AI Agents with Tools, Memory, and Learning"},"content":{"rendered":"

Researchers at Stanford University and Lambda Labs, have published the research paper for OpenJarvis<\/a>, an open-source framework that runs inference, agents, memory, and learning entirely on-device. <\/p>\n

The open-weight models configured through OpenJarvis land within 3.2 percentage points of the best cloud model on average, at roughly 800\u00d7 lower marginal API cost per query and roughly 4\u00d7 lower latency under the research\u2019s benchmark protocol. This research work builds on the research team\u2019s earlier Intelligence Per Watt<\/em> study<\/a>, which reported that local models already handle 88.7% of single-turn chat and reasoning queries at interactive latency, with intelligence efficiency improving 5.3\u00d7 from 2023 to 2025.<\/p>\n

Model Overview & Access<\/strong><\/h2>\n

OpenJarvis is not a single model. It is a framework that composes any supported model with a configurable agent stack, evaluated across 11 local models from four families.<\/p>\n

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Property<\/th>\nValue<\/th>\n<\/tr>\n<\/thead>\n
License<\/strong><\/td>\nApache 2.0<\/td>\n<\/tr>\n
Framework release<\/strong><\/td>\nMarch 12, 2026<\/td>\n<\/tr>\n
Paper<\/strong><\/td>\narXiv:2605.17172 (posted May 16, 2026)<\/td>\n<\/tr>\n
Repository<\/strong><\/td>\ngithub.com\/open-jarvis\/OpenJarvis<\/td>\n<\/tr>\n
Stars \/ forks<\/strong><\/td>\n~5.4k \/ ~1.2k (June 2026)<\/td>\n<\/tr>\n
Languages<\/strong><\/td>\nPython (~83%), Rust (~9%), TypeScript (~7%)<\/td>\n<\/tr>\n
Evaluated models<\/strong><\/td>\n11 local models across 4 families: Qwen3.5, Gemma4, Nemotron, Granite<\/td>\n<\/tr>\n
Cloud baselines<\/strong><\/td>\nClaude Opus 4.6, GPT-5.4, Gemini 3.1 Pro<\/td>\n<\/tr>\n
Supported engines<\/strong><\/td>\nOllama, vLLM, SGLang, llama.cpp, Apple Foundation Models, Exo (among others)<\/td>\n<\/tr>\n
Context window<\/strong><\/td>\nModel-dependent<\/td>\n<\/tr>\n
Installation<\/strong><\/td>\nSingle command; ~3 minutes on broadband<\/td>\n<\/tr>\n
Hardware<\/strong><\/td>\nTested on 7 platforms, from Mac Mini M4 to NVIDIA DGX Spark<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/figure>\n

Architecture: Five Primitives and a Spec<\/strong><\/h2>\n

OpenJarvis decomposes a personal AI system into five typed primitives, composed through a single declarative configuration object called a spec<\/strong>.<\/p>\n