Is OpenJarvis really from Stanford?

Yes. The README says it is developed at Hazy Research and the Scaling Intelligence Lab at Stanford SAIL, and the paper is arXiv 2605.17172. The project site is openjarvis.stanford.edu.

Yes. It is open source under the Apache-2.0 license. With a local model there are no API fees. You only pay if you choose to connect a paid cloud model.

By default the server only listens on your own computer (127.0.0.1). The risk comes if you bind it to 0.0.0.0 or use the server profile on a shared network. Keep it on 127.0.0.1, and create an API key with jarvis auth create-key before you expose it anywhere.

Does it work on Windows?

Yes. There is a Windows .exe desktop installer on the releases page, a PowerShell one-liner for native Windows, and the bash installer works inside WSL2.

Can I message it from my phone?

Yes. The repo includes Telegram, WhatsApp and iMessage channel connectors, so you can text your assistant and it answers from your own machine.

Getting started

Stanford Just Released a Free Jarvis: How to Install OpenJarvis

6 minute readUpdated September 2026Explore more

TL;DR

OpenJarvis is a free, open-source personal AI assistant from Stanford's Hazy Research and Scaling Intelligence Lab. It runs a local model on your own computer, so everyday use costs $0 in API fees. Setup is one command in your terminal. Then pick the morning briefing preset, log in with Google, and type jarvis digest to hear your email and calendar read out loud. You can also talk to it from Telegram, WhatsApp or iMessage, and run a research agent that cites its sources. Before you connect anything, read the security section: keep the server on 127.0.0.1 unless you have set an API key.

Learn Claude Code. Earn income. Only $9.

👉 https://www.skool.com/claudecodeclub

What OpenJarvis is

OpenJarvis is a personal AI assistant that runs on your own device instead of someone else's server. It was built at Stanford by Hazy Research and the Scaling Intelligence Lab (part of Stanford SAIL), it is open source under the Apache-2.0 license, and the research behind it is published as arXiv paper 2605.17172. The repo has passed 10,000+ stars on GitHub. We checked on September 29, 2026.

open-jarvis/OpenJarvis

10,000+ stars · Apache-2.0 · Personal AI, on personal devices

The idea is simple. Use a local model by default and only call a cloud model when you choose to. The project site is https://openjarvis.stanford.edu/ and the paper is at https://arxiv.org/abs/2605.17172

Security first: keep it on 127.0.0.1

OpenJarvis can run a web server (jarvis serve) so other apps can talk to it. Out of the box it only listens on your own computer (127.0.0.1, set in src/openjarvis/core/config.py line 1165). Two things open it to every device on your network: running jarvis serve --host 0.0.0.0 (shown in the project's CLI docs), or picking the server security profile, which sets host 0.0.0.0 (config.py line 1497). On a shared network like an office, a café or a dorm, anyone on that network could then reach your assistant. Our rule: leave the host on 127.0.0.1. If you really need it on your network, create an API key first with jarvis auth create-key (or set the OPENJARVIS_API_KEY environment variable). This matters most once you add a cloud API key, because an open server can spend that key. The project refuses to start on a non-local address without a key (check_bind_safety in src/openjarvis/server/auth_middleware.py), which helps, but don't rely on it alone.

How to install OpenJarvis

Option 1: the one-command install (the one from the reel). Paste this single line into your terminal on a Mac, Linux or WSL2. The script is hosted on the project's own GitHub Pages site (open-jarvis.github.io), not on GitHub itself. Open the URL in your browser and read it before you pipe it into bash.

bashcurl -fsSL https://open-jarvis.github.io/OpenJarvis/install.sh | bash

Per the README, this installer sets up uv, a Python environment, Ollama and a small starter model, in about 3 minutes on broadband. Then type jarvis to start, and jarvis doctor to check the status.

Option 2: the desktop app. Prefer not to use the terminal? Go to the releases page and download the file for your computer: .dmg for Mac, .exe for Windows, .deb for Ubuntu or Debian (there are also .rpm and .AppImage builds).

OpenJarvis desktop app releases

Mac .dmg · Windows .exe · Linux .deb, .rpm, .AppImage

Option 3: from source, if you want to read the code first. These are the commands from the project's installation docs. You need uv (https://docs.astral.sh/uv/) and Rust installed, plus a running local model backend such as Ollama (https://ollama.com/).

bashgit clone https://github.com/open-jarvis/OpenJarvis.git
cd OpenJarvis
uv sync
uv run maturin develop -m rust/crates/openjarvis-python/Cargo.toml

Then check it works with uv run jarvis --version and start chatting with uv run jarvis.

What you can do with it

1. A spoken morning briefing. The morning digest preset reads your email, calendar, health data and news, then plays you a short audio briefing. These are the README commands for a Mac:

bashjarvis init --preset morning-digest-mac --force
jarvis connect gdrive   # one sign-in covers Gmail, Calendar and Tasks
jarvis digest --fresh   # make and play your first briefing

On Linux, use morning-digest-linux instead. The full guide is at https://open-jarvis.github.io/OpenJarvis/user-guide/morning-digest/

2. Talk to it from your phone. The repo includes channel connectors for Telegram, WhatsApp and iMessage (src/openjarvis/channels/telegram.py, whatsapp.py and imessage_daemon.py). You message it the way you'd message a friend, and it replies from your own computer.

3. Eight built-in agents. The README lists eight, including deep_research (multi-hop research with citations across the web and your local docs), morning_digest, a scheduled monitor_operative, and native_openhands, which writes and runs Python code.

bashjarvis init --preset deep-research --force

What it costs

The software is free. If you run a local model, answers are generated on your own computer, so there is no per-message API bill. That is $0 in API fees, but your computer does the work, so bigger models need more memory. You can add cloud models later, and that is when the security callout above matters most.

Quick start in 10 minutes

1Paste the one-command install into your terminal (or grab the desktop app).
2Start it and ask a simple question to make sure the local model answers.
3Pick the morning digest preset, log in with Google (jarvis connect gdrive), and run jarvis digest --fresh.
4Leave the server host on 127.0.0.1. Only change it after running jarvis auth create-key.

Learn Claude Code. Earn income. Only $9.

👉 https://www.skool.com/claudecodeclub

Common questions

Is OpenJarvis really from Stanford?
Yes. The README says it is developed at Hazy Research and the Scaling Intelligence Lab at Stanford SAIL, and the paper is arXiv 2605.17172. The project site is openjarvis.stanford.edu.
Is OpenJarvis free?
Yes. It is open source under the Apache-2.0 license. With a local model there are no API fees. You only pay if you choose to connect a paid cloud model.
Is it safe to run?
By default the server only listens on your own computer (127.0.0.1). The risk comes if you bind it to 0.0.0.0 or use the server profile on a shared network. Keep it on 127.0.0.1, and create an API key with jarvis auth create-key before you expose it anywhere.
Does it work on Windows?
Yes. There is a Windows .exe desktop installer on the releases page, a PowerShell one-liner for native Windows, and the bash installer works inside WSL2.
Can I message it from my phone?
Yes. The repo includes Telegram, WhatsApp and iMessage channel connectors, so you can text your assistant and it answers from your own machine.

Keep going

Build your own AI assistant with 8,000+ Claude Code builders for $9/month

Get 650+ plug-and-play skills, MCPs & prompts, plus 8,000+ members - $9/mo, cancel anytime.

Join the Club