Version 1.0 of OpenJarvis, an open-source framework for assembling personal AI agents that run entirely on your own computer, is now available[1]. It was built by two Stanford University labs, Hazy Research and Scaling Intelligence, and ships with built-in support for Ollama, a tool for running large language models (LLMs) locally[1]. Its defining trait is making "local-first" the default, so work is completed on your own hardware rather than leaning entirely on the cloud.
Rethinking the Cloud-First Assumption
OpenJarvis is an open-source framework, a foundational structure for development, for building personal AI agents that run on your own hardware[1]. Stanford University's Hazy Research and Scaling Intelligence labs built it as part of their "Intelligence Per Watt" research into efficient local AI[1].
The team points to a gap: local models can already handle most everyday chat and reasoning, yet most personal AI still sends every request to the cloud[1]. OpenJarvis flips that assumption and makes local-first the default behavior[1]. Models run on your own machine, and using the cloud is entirely optional[1]. On top of that, it tracks energy, cost, and latency, the time it takes to respond, right alongside accuracy[1].
When processing stays on your own device, your input does not have to travel to an external server, which brings benefits for privacy and communication costs as well. In version 1.0, Ollama support is built in as the foundation for this local execution[1].
Installation and Choosing a Model
To get started, you first set up Ollama, which runs models locally[1]. Ollama is distributed for macOS, Windows, and Linux[1].
Next you install OpenJarvis itself. On macOS and Linux, running the install script sets up everything you need and automatically detects an existing Ollama installation[1].
curl -fsSL https://open-jarvis.github.io/OpenJarvis/install.sh | bash
On Windows, you run that command inside WSL2, a mechanism for running a Linux environment on Windows, or install the desktop app[1]. Once everything is ready, running the command "jarvis" starts it up[1].
You can choose the model yourself[1]. The install script sets up a starter model so you can begin right away, but you can also pull and use any model through Ollama[1].
jarvis model pull qwen3.5:35b
jarvis ask -m qwen3.5:35b "Your prompt"
To fix the default model, you add it to the configuration file "~/.openjarvis/config.toml" as follows[1].
[intelligence]
default_model = "qwen3.5:35b"
preferred_engine = "ollama"
Built-in Agents Ready to Run
OpenJarvis ships with presets, ready-made combinations of settings[1]. Each one bundles the agent, execution engine, and tools needed for a particular use case[1].
The first is a morning briefing. It generates a morning digest agent that uses your calendar, email, and the day's news to summarize the key points[1].
jarvis init --preset morning-digest-mac
jarvis connect gdrive
jarvis digest --fresh
The second is research across files. It searches across information on the web and your local documents and returns an answer with citations[1].
jarvis init --preset deep-research
jarvis memory index ./docs/
jarvis ask "Summarize all emails about Project X"
The third is a local coding agent. It writes and runs Python on your own machine to get tasks done[1].
jarvis init --preset code-assistant
What all of these presets share is that they run in your own environment without staying constantly connected to the cloud. The design lets you pick an initial configuration to match your use case and swap in a different model when you need to.
Summary
OpenJarvis is an open-source framework for running personal AI agents on your own hardware, and from version 1.0 it includes Ollama support out of the box. Against the backdrop of the Stanford labs' "Intelligence Per Watt" research, it shifts work that used to be left to the cloud toward local-first, and it records not only accuracy but also energy, cost, and latency. Presets for a morning digest, research across files, and local coding assistance are included, so you can try it on your own machine the moment you install it.
