# Jarvis — AI voice assistant and agent for Mac > Jarvis is a free, open-source, on-device voice assistant for Mac (4x faster dictation plus voice control of your apps). Jarvis 2.0 (private beta) adds persistent memory, routines that run while you sleep, and action across your connected apps. These docs, "The Console," teach how to use it from your first word to autopilot. Docs home: https://jarvis.ceo/docs ## Start here - [Your first 5 minutes](https://jarvis.ceo/docs/start): Install Jarvis, grant two permissions, and say your first word. Five minutes from download to dictating into any app on your Mac. - [L1 · Your first word](https://jarvis.ceo/docs/path/first-word): The first rung of the Jarvis mastery path. Hold, speak, release, and understand exactly what just happened on-device. - [L2 · Flow](https://jarvis.ceo/docs/path/flow): The second rung: stop reaching for the keyboard. Punctuation, formatting, and edit-by-voice make dictation faster than typing. - [L3 · Control](https://jarvis.ceo/docs/path/control): The third rung: stop clicking. Open apps, switch windows, and trigger actions with your voice. - [L4 · Memory](https://jarvis.ceo/docs/path/memory): The fourth rung: tell Jarvis something once and it remembers. This is where the assistant becomes yours. - [L5 · Autopilot](https://jarvis.ceo/docs/path/autopilot): The final rung: routines that run on their own. Jarvis does the recurring work before you ask. ## Talk and type - [Getting started](https://jarvis.ceo/docs/dictation/getting-started): How dictation works in Jarvis: push-to-talk, on-device transcription, punctuation, and the editing commands that make voice faster than typing. - [Commands cheatsheet](https://jarvis.ceo/docs/dictation/commands-cheatsheet): A reference of voice commands that work on macOS today with Jarvis: dictation editing, app control, browser, mail, Slack, and coding. Hold the hotkey, say the phrase, release. - [Troubleshooting](https://jarvis.ceo/docs/dictation/troubleshooting): Nothing typing? Mic not detected? The two permissions Jarvis needs and how to fix the most common dictation problems on macOS. ## Operate your Mac - [Voice control basics](https://jarvis.ceo/docs/control/getting-started): Two ways to drive your Mac with Jarvis: talk to the orb, or type in chat. Open apps, look at your screen, and trigger actions by voice. - [Ask about your screen](https://jarvis.ceo/docs/control/screen-vision): Jarvis can see what is on your screen and answer questions about it. Ask what an error means, summarize a document, or pull a number off a dashboard. - [Multi-step chains](https://jarvis.ceo/docs/control/chains): Ask for something that takes several steps, and Jarvis plans it, does the work, and reports back. How chaining and delegation work. ## The agent that knows you - [What is Jarvis](https://jarvis.ceo/docs/jarvis-2-0/overview): Jarvis is the agent that knows you: persistent memory, routines that run while you sleep, and action across your apps. In early access. - [Memory](https://jarvis.ceo/docs/jarvis-2-0/memory): How Jarvis remembers: the people, projects, places, and preferences it keeps on your Mac, how to add or forget them, and why it stays private. - [Routines](https://jarvis.ceo/docs/jarvis-2-0/routines): Turn any recurring task into a routine that runs on its own, including while you sleep. How to create one, where results land, and the one setup step to know. ## How it works - [How Jarvis thinks](https://jarvis.ceo/docs/concepts/how-jarvis-thinks): The mental model: a fast voice layer that answers and routes instantly, and a stronger agent that does the heavy work, in the foreground or quietly in the background. - [How Jarvis learns](https://jarvis.ceo/docs/concepts/how-jarvis-learns): Jarvis gets better the more you use it: it learns your writing voice, notices habits worth automating, and refines from your feedback. All on your Mac. ## Recipes - [Triage your inbox before standup](https://jarvis.ceo/docs/recipes/triage-inbox-before-standup): A day-in-the-life recipe: have Jarvis read overnight email, flag what needs you, and draft replies, so you walk into standup already on top of it. - [Your morning briefing](https://jarvis.ceo/docs/recipes/morning-briefing): Wake up to a recap of what changed overnight: a routine that gathers your projects, mail, and calendar and sends one briefing to your chat before you start. - [Clear Slack by voice](https://jarvis.ceo/docs/recipes/clear-slack-by-voice): Have Jarvis summarize your unread Slack, surface what needs a reply, and draft responses you approve. Get to inbox zero without typing. - [Research and summarize](https://jarvis.ceo/docs/recipes/research-and-summarize): Hand Jarvis an open question and let it do the legwork: it plans the steps, searches, and returns a short answer with the trade-offs, not ten tabs. - [Schedule a meeting by voice](https://jarvis.ceo/docs/recipes/schedule-a-meeting): Find a time, create the event, and invite people without opening your calendar. Jarvis checks your availability and books it. - [Summarize a document](https://jarvis.ceo/docs/recipes/summarize-a-document): Point Jarvis at a long document on your screen or in a connected app and get the gist, the decisions, and the action items in seconds. - [Prep for a meeting](https://jarvis.ceo/docs/recipes/prep-for-a-meeting): Walk in ready: ask Jarvis to pull who you are meeting, the latest threads, and the agenda, so you are briefed in a minute instead of digging through tabs. - [End-of-day shutdown](https://jarvis.ceo/docs/recipes/end-of-day-shutdown): Close the day clean: a routine that recaps what you did, drafts tomorrow top three, and surfaces anything you dropped, so work does not follow you home. - [Voice code-review in Cursor](https://jarvis.ceo/docs/recipes/voice-code-review-in-cursor): Open Cursor, dictate prompts hands-free, and have Jarvis delegate the actual review to a coding agent, then show you the diff. An honest, working flow. - [Dictate a long email](https://jarvis.ceo/docs/recipes/dictate-a-long-email): A simple recipe: write a full email by voice, with punctuation and paragraph breaks, then fix it without touching the keyboard. ## Switching to Jarvis - [Coming from Wispr Flow](https://jarvis.ceo/docs/coming-from/wispr-flow): Switching from Wispr Flow to Jarvis: what feels the same, what is different, and how to set up dictation in two minutes. Free, open source, and on-device. - [Coming from Superwhisper](https://jarvis.ceo/docs/coming-from/superwhisper): Switching from Superwhisper to Jarvis: you keep local, private transcription and gain voice control of your Mac plus an agent that remembers and acts. - [Coming from Apple Dictation](https://jarvis.ceo/docs/coming-from/apple-dictation): Switching from built-in Apple Dictation to Jarvis: faster, more accurate on technical English, with no phrase limit, plus voice control and an agent. ## Extend with MCP - [Extend Jarvis with MCP](https://jarvis.ceo/docs/mcp/overview): MCP (Model Context Protocol) is the open standard for giving an AI assistant new tools. Jarvis speaks it both ways: connect any MCP server, and expose Jarvis to your other agents. - [Connect an MCP server](https://jarvis.ceo/docs/mcp/connect-a-server): Three ways to add an MCP server to Jarvis: a one-click catalog, automatic discovery of servers you already use in Cursor or Claude Code, and a custom-server form. Supports stdio, HTTP, and SSE. - [Use Jarvis from your tools](https://jarvis.ceo/docs/mcp/use-jarvis-from-your-tools): Jarvis publishes an MCP server (jarvis-memory) that exposes your memory and connected apps, so a coding agent like Claude Code, Cursor, or Codex can use what Jarvis knows. - [Build a tool for Jarvis](https://jarvis.ceo/docs/mcp/build-a-server): If the tool you want does not exist, build it. Any server that speaks MCP works with Jarvis. How to scaffold one and point Jarvis at it. ## The Pantry - [Command reference](https://jarvis.ceo/docs/pantry/commands): The canonical reference of what Jarvis can do, grouped by capability, with what each one needs. For the friendly version, see the cheatsheet. - [Transcription and offline](https://jarvis.ceo/docs/pantry/transcription-and-offline): How Jarvis turns speech into text: on-device models that work fully offline, with an optional cloud model you bring your own key for. Accuracy, speed, and privacy. - [Hotkeys](https://jarvis.ceo/docs/pantry/hotkeys): Every Jarvis keyboard shortcut: push-to-talk, hands-free toggle, cancel, and how to change the trigger key to whatever fits your hands. - [Privacy and your data](https://jarvis.ceo/docs/pantry/privacy): How Jarvis keeps your data private: on-device transcription, local memory, no account to start, and optional cloud features that use your own keys. - [Generative UI cards](https://jarvis.ceo/docs/pantry/card-types): Jarvis answers by drawing the right interface, not a wall of text. The card types it generates: scorecards, charts, timelines, checklists, tables, code, and diffs. - [Connections](https://jarvis.ceo/docs/pantry/connections): The apps and services Jarvis can connect to, what each one unlocks, and what to expect on setup. Plus MCP for everything else. - [Skills](https://jarvis.ceo/docs/pantry/skills): Skills are reusable workflows Jarvis can run. It ships with built-in skills, learns new ones from what you do repeatedly, and reaches further through connections and MCP. ## Product - [Download Jarvis (free, open source)](https://jarvis.ceo/download) - [Jarvis 2.0 — join the waitlist](https://jarvis.ceo/jarvis-2-0) - [GitHub: jarvis-ai-assistant](https://github.com/akshayaggarwal99/jarvis-ai-assistant) ## Optional - [llms-full.txt](https://jarvis.ceo/llms-full.txt): the complete docs as a single file.