Interactive demo of the app. It opens on the front page in voice mode. The person says the login test is failing and asks for an agent to fix it; the assistant finds the failed run on GitHub and hands the job to an agent called Scout, then says so out loud. The demo then moves to the Agentic IDE, where three coding agents work side by side in one project: Scout, on Claude Code, fixes the test and opens a pull request; Atlas, on Codex, updates dependencies; Quill, on Gemini CLI, drafts the release notes. Two rows in the sidebar, New chat and Agentic IDE, move between the two scenes.

Everything Jarvis connects

  • AirtableOllamaGitHubAntigravityOpenRouterGmailAWS
  • Google DriveCanvaSpotifyGroqClickUpSupabaseHome Assistant
  • TwilioNeonElevenLabsxAI GrokNotionFly.ioxAI Grok

Plugins connect Jarvis to the tools you already use. Sign in once, and it can read your mail, find a file or start the music. When a connection stops working, it tells you which one and why.

See every plugin

A demo of the app's Plugins section. The whole catalog: twenty-four services, two of them connected here. Searching for “google” narrows the list to Gmail, Google Drive and Google Calendar. Connecting Gmail. Gmail is connected and answering. A warning above the list: Spotify needs reconnecting because the provider withdrew the authorization.

Skills are what Jarvis knows how to do. Each one is a plain text file with a switch: a few lines saying what it is for, and what sets it off — a spoken phrase, a shortcut, or the clock. The ones it ships with and the ones you write sit in the same list.

See how a skill is written

A demo of the app's Skills section. A list of skills. Each is a file with a switch, and the one you wrote sits in the same list as the ones that shipped. Searching for “summary” narrows the list to nine, and every hit says which field matched. Opening morning-routine shows what it is for, and its status, version and category, read out of the file itself. Its triggers: two spoken phrases and one clock entry at seven in the morning. Those three lines are the whole reason the briefing starts. Filtering to “Mine” leaves the one skill written here, three-bullet-brief. Click it to read its card.

It drives the command-line tools you already have. Ask in plain language; it picks the tool, runs the command, and shows you exactly what it ran and why that was allowed. Every tool carries its own list of commands that are refused before they start.

See every command-line tool

A demo of the app's CLIs section. The whole catalog of command-line tools Jarvis can drive: twenty-two of them, three connected here. Opening the GitHub CLI shows the binary it runs and the exact commands it uses to check it and sign in. Each tool carries its own blocked and allowed command patterns, so deleting a repository is refused before it is ever run. In the test hub you ask in plain language: show my open pull requests. The answer comes with its evidence: which tool was picked, which risk tier the gate resolved, the exit code and the exact command that ran.

Agents

A team that works while you don't.

Each agent has its own chat, model, browser and routines. Jarvis leads, they talk to each other, and anything that sends waits for your OK. Click around — it's the real screen, with sample data.

What changes

Most assistants talk back.
This one does the thing.

It runs on your own machine and answers to a name you pick. Small things it does itself — opens the app, changes the setting, takes the mouse. Real work goes to a coding agent in an isolated copy of your files and comes back finished.

A home office in hard daylight, printed as fine dark dots on paper white, seen over the shoulder of a person hunched at the desk with both hands on the keyboard. The wide monitor facing the camera shows a code editor buried under a dozen overlapping panels, its tab row crammed and its terminal full. A mug, a plant and a slipped stack of papers sit at the left, a bright window behind.The same desk, the same room, the same mug and plant, at night, from the same place over the shoulder. The dots have inverted to light on black. The person has sat back with both hands away from the keyboard and mouse, head up, speaking into a microphone on a boom arm. The same editor is on the monitor, now down to a single tidy column of code in colour with empty space around it.

Hey Jarvis, the tests are red. Find out why and fix it.

Hands on the keyboard. You find the window, the tab and the right button, then read the result yourself.
ProvidersBrain, voice in, voice out
32
PlatformsWindows, macOS, Linux
3
Spoken replyRealtime model
< 1 s
ChannelsOne brain, one memory
4

In the open

160GitHub stars


Live from GitHub, counted again every time this page loads.

The globe marks 28 places, covering 32 of 160 stargazers. Only public locations we can resolve are shown. New stars update the map automatically; processing can take a few minutes. Locations checked .

  • Brazil: 3 stargazers
  • India: 3 stargazers
  • Auckland, New Zealand: 1 stargazer
  • Belgium: 1 stargazer
  • Berlin, Germany: 1 stargazer
  • Bhubaneswar, India: 1 stargazer
  • Calgary, Canada: 1 stargazer
  • Chennai, India: 1 stargazer
  • Chicago, United States: 1 stargazer
  • Dhaka, Bangladesh: 1 stargazer
  • Dominican Republic: 1 stargazer
  • Indianapolis, United States: 1 stargazer
  • Los Angeles, United States of America: 1 stargazer
  • Minnesota, United States: 1 stargazer
  • Moldova: 1 stargazer
  • Munich, Germany: 1 stargazer
  • Oostanay, Kazakhstan: 1 stargazer
  • Pakistan: 1 stargazer
  • Paris, France: 1 stargazer
  • Qazvin, Iran: 1 stargazer
  • Romania: 1 stargazer
  • Rome, Italy: 1 stargazer
  • Slovakia: 1 stargazer
  • South America: 1 stargazer
  • Surabaya, Indonesia: 1 stargazer
  • Sydney, Australia: 1 stargazer
  • United States: 1 stargazer
  • Vienna, Austria: 1 stargazer

On camera

This is being built in public.You can watch how it happens.

Development happens on camera, and the code lands in a public repository anyone can read. Everything published so far is below.

All things you have to know about personal Jarvis V2

Watch on YouTube

Nothing loads from YouTube until you press play. Stars and members update live; subscribers and followers were read on 29 Aug 2026.

Open source

Read it before you run it.Every line is on GitHub.

This is the repository the install command pulls from. Open it yourself — nothing here sits behind a signup.

A walkthrough of the PersonalJarvis/PersonalJarvis repository on GitHub: the root listing, the application package, the install directory, and the first lines of install.ps1 — the script the install command runs. The repository is public and Apache-2.0-licensed.

Getting started

How to install Jarvis

  1. 1

    One command

    Windows, macOS or Linux, one line. You need Python 3.11 or newer and Git; the installer checks for both and stops with a download link if either is missing. It asks nothing else, builds its own environment, fetches the voice models and opens the app.

    TerminalPowerShellShell

    curl -fsSL https://raw.githubusercontent.com/PersonalJarvis/PersonalJarvis/main/install/install.sh | bash

    irm https://raw.githubusercontent.com/PersonalJarvis/PersonalJarvis/main/install/install.ps1 | iex

    curl -fsSL https://raw.githubusercontent.com/PersonalJarvis/PersonalJarvis/main/install/install.sh | bash

  2. 2

    Pick a wake word

    The app walks you through language, wake word and keys once. Nothing is preset — you type any phrase you like, and the assistant takes its name from it.

  3. 3

    Add a key, or skip it

    In the same setup you hand over a provider key or a subscription login — Gemini, Claude, OpenAI or OpenRouter. It goes into your operating system's credential manager, never into a file. Every tier also has a local option that needs no key at all, so this step is skippable.

  4. 4

    Say it

    jarvis opens the desktop window with voice and the Orb overlay. On a server, jarvis serve runs the same assistant in a browser tab, microphone and all.

Talk to your computer.
It does the thing.

Apache 2.0, running on your own machine. Bring the AI subscription you already pay for, or an API key, or nothing at all and keep it offline.

Install Personal Jarvis