Jarvis on GitHub: 4 Open-Source Jarvis Projects (Ours Isn't One).

Disclosure: This page is published by the team behind Jarvis (getjarvis.eu), a paid desktop assistant for macOS, Windows, and Linux. Our product is closed-source, and we are not affiliated with any of the open-source projects listed below.

People search for an open Jarvis, an open-source Jarvis, or a Jarvis GitHub repo for the same reason: they want an assistant they can read, self-host, and change. That is a legitimate thing to want, and we're not going to try to talk you out of it. If you came for a repository, the projects below are real and cloneable — links and commands first, opinions after. Then, because we build a closed, cloud-backed assistant in the same space, we'll lay out exactly what that trade-off buys and what it costs, so you can decide with real information instead of marketing.

One honesty note before the list: we have not audited these codebases. Star counts, requirements, and feature claims below come from each project's own GitHub page and documentation as we read them in July 2026 — verify against the repo before you build on any of them.

A research framework for personal AI agents that run on your own hardware, with cloud APIs optional rather than required. Per the project's docs, the quickstart script installs dependencies, launches Ollama with a local model, and opens a web interface at localhost:5173. The project describes itself as developed at Stanford (Hazy Research and the Scaling Intelligence Lab) and lists sponsors including the Laude Institute, Google Cloud, Lambda, and IBM Research — unusually strong institutional backing for a project in this space, if it holds up, and probably why it currently tops this search.

Honest limits: The quickstart still assumes git, a terminal, and hardware capable of running local models. Its headline claim — that local models already handle 88.7% of single-turn chat and reasoning queries — comes from the project's own paper, so treat it as a thesis the project exists to prove, not an independently verified fact.

Around 17k GitHub stars as of July 2026, with years of sustained development. Per its documentation, it's built on Node.js (version 24 or newer), structured around tools, context, memory, and agent-style task execution, and explicitly privacy-first: the core runs on your machine.

Honest limits: At the time of writing, the new 2.0 architecture is a developer preview on the develop branch; the stable branch is the older design, so you're choosing between mature-but-dated and modern-but-moving. It's a skills-and-workflows assistant — it does not watch your screen or operate your applications for you.

Roughly 25k stars as of July 2026. This is the code behind a well-known 2023 research paper: a large language model acts as a controller that plans a task, selects expert models from Hugging Face, executes them, and composes the result. Genuinely fascinating architecture, and the repo has continued to ship research follow-ups.

Honest limits: It's a research system, not a daily-driver assistant. The repo's default configuration asks for Ubuntu, a GPU with at least 24 GB of VRAM, and roughly 284 GB of disk for model downloads (a lite mode is far lighter but far less capable). Expect research-grade rough edges.

Open source Jarvis